Journal of Science Engineering Technology and Management Science

Medical imaging plays a vital role in the early detection and treatment monitoring of lung cancer. Common imaging techniques such as chest X-rays, computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), and other molecular imaging methods have long been used to diagnose lung cancer. Despite their widespread use, these modalities face certain limitations particularly their lack of automated capabilities for accurately classifying cancerous tissues. This challenge becomes even more critical in patients with overlapping or co-existing conditions, where misdiagnosis can delay treatment. To address this gap, there is a growing demand for diagnostic tools that are not only highly sensitive but also precise in identifying early-stage lung cancer. In recent years, deep learning has emerged as a transformative technology in medical imaging, offering robust capabilities for analyzing both visual and textural data. Leveraging this potential, the present study introduces an enhanced convolutional neural network (CNN) model specifically designed for detecting and classifying lung cancer from chest CT scans. The proposed CNN architecture is capable of categorizing images into four distinct classes: adenocarcinoma, large cell carcinoma, squamous cell carcinoma, and normal tissue. When benchmarked against traditional machine learning approaches specifically the Naïve Bayes Classifier (NBC) the deep CNN model demonstrated superior performance and higher classification accuracy. Evaluation metrics further underscore the model’s effectiveness, suggesting that it can significantly aid clinicians in making more accurate and timely diagnoses, ultimately improving patient outcomes

Keywords : Lung Cancer Detection, Early Diagnosis, Deep Learning in Medical Imaging, Convolutional Neural Network (CNN).

Author : Vipul G Gajjar, Kamalesh V. N.,Kavitha Rani Balmuri

Title : CANCER CLASSIFICATION UTILIZING MEDICAL IMAGES

Volume/Issue : 2025;02(7S)

Page No : 1-10

The rapid expansion of the Internet of Things (IoT) has transformed numerous industries by enabling real-time data exchange and automated operations. However, this growth brings significant challenges in maintaining robust security and efficient system performance. Recent projections indicate that the number of IoT devices will surpass 75 billion by 2025, resulting in a massive, interconnected network vulnerable to security threats and maintenance difficulties. Traditional security mechanisms, which rely on manual monitoring and fixed threshold-based detection, are increasingly inadequate in dynamic IoT environments. These conventional approaches often fail to detect new and evolving threats in real time and struggle with predictive maintenance, leading to higher downtime and degraded system performance. Additionally, the heterogeneous nature of IoT devices and the massive data they generate make it difficult for legacy systems to respond quickly and effectively. These limitations highlight the urgent need for more advanced solutions. This research explores the application of deep learning (DL) models to meet these challenges by focusing on security threat detection, performance forecasting, and predictive maintenance in IoT ecosystems. DL techniques offer the ability to process large-scale data, identify anomalies indicative of security issues, and forecast maintenance requirements before failures occur. The strength of this approach lies in its capacity to enhance IoT system reliability and security, reduce downtime risk, and improve performance management. Ultimately, this research aims to enable a shift from reactive to proactive management of IoT systems, supporting their scalability and complexity while ensuring secure and efficient operation for critical applications

Keywords : Internet of Things (IoT),

Author : Dr. Pulime Satyanarayana, Chepuri Bharath Kumar, Pagidala Bharath Kumar, Godugu Elishan

Title : KNOWLEDGE-AWARE IOT MAINTENANCE PREDICTION USING DEEP LEARNING ON STACK OVERFLOW SECURITY DATA

Volume/Issue : 2025;02(7S)

Page No : 11-19

Adverse drug reactions pose significant risks in clinical settings, especially when drug side effects are overlooked during early prescription stages. To mitigate such risks, this study focuses on enhancing drug side effect prediction using machine learning techniques integrated with Explainable AI (XAI) for medical health applications. The core objective is to develop an intelligent, interpretable system that not only predicts potential side effects but also provides transparency into the decision-making process, fostering trust in healthcare professionals a comprehensive dataset comprising drug attributes, side effect profiles, and associated clinical features was used for model training and evaluation. Initial experimentation was conducted using various baseline classifiers including Ridge Classifier, Linear Support Vector Machine (SVM), Logistic Regression, and Multinomial Naïve Bayes. These models served as benchmarks for performance in terms of accuracy, precision, recall, and F1-score. Extensive Exploratory Data Analysis (EDA) was performed to uncover patterns, correlations, and imbalances in the dataset, aiding in informed feature selection and preprocessing. To improve prediction accuracy and enable complex pattern recognition, a Multi-Layer Perceptron (MLP) Classifier was proposed as the advanced model. The MLP model, being a deep learning algorithm, demonstrated superior performance in capturing nonlinear relationships among features that traditional models often fail to detect. This makes the solution viable for real-world deployment in clinical decision support systems (CDSS), ensuring safer drug administration and better patient outcomes the project contributes to the field of medical AI by delivering a high-performing, interpretable, and reliable solution for drug side effect prediction, bridging the gap between complex AI models and practical healthcare applications

Keywords : Explainable Artificial Intelligence, Drug Side Effect Prediction, Healthcare Risk Mitigation, Pharmacovigilance

Author : Dr. S. Sankar Ganesh, Gajji Abhi Sai, Velagala Ashruth Rajesh, Pusa Prashanth

Title : EXPLAINABLE AI FOR DRUG SIDE EFFECT PREDICTION: A TRANSPARENT APPROACH FOR SAFER MEDICAL DECISIONMAKING

Volume/Issue : 2025;02(7S)

Page No : 20-29

The surge in online activity has significantly increased cybersecurity threats, as users often engage in behaviors that expose them to risks such as data breaches, identity theft, and other cyber incidents. While conventional cybersecurity tools such as firewalls, antivirus software, and intrusion detection systems offer a degree of protection, they tend to be reactive and are limited by their reliance on static thresholds and signature-based detection. These limitations reduce their effectiveness in detecting subtle or emerging behavioral threats. The project introduces a Behavioral Risk Classifier, a machine learning-based system designed to analyze and categorize user behavior as either ‘safe’ or ‘risky.’ It leverages rich behavioral features such as device type, geolocation, network type, social media usage, and e-safety awareness scores to identify potential risks. The preprocessing module cleans and transforms the data, decomposes timestamps for detailed temporal analysis, encodes categorical features, and visualizes risk category distributions. To address class imbalance, the system applies the Synthetic Minority Over-sampling Technique (SMOTE), promoting balanced learning across both classes. The dataset is further refined using StandardScaler for normalization and Principal Component Analysis (PCA) for dimensionality reduction. Two predictive models are developed: a Gradient Boosting Classifier (GBC) for capturing complex data patterns and a hybrid model that integrates a Deep Neural Network (DNN) for feature extraction with a Random Forest Classifier (RFC) for final prediction. This hybrid setup combines the deep learning capabilities of DNNs with the interpretability and robustness of RFCs. The system also features performance evaluation tools, detailed classification reports, and a real-time prediction module for continuous risk assessment. By proactively analyzing user behavior and adapting to emerging threats, this solution offers a powerful and accurate approach to cybersecurity, helping organizations better safeguard their digital environments.

Keywords : Behavioral Risk Classification, Cybersecurity, E-Safety Awareness, Behavior-Based Security

Author : Dr. B. Rama, Suroju Dinesh Karthik, Tummala Likith, Suthoju Sai Rohith

Title : ADAPTIVE CYBER DEFENSE: CLASSIFYING BEHAVIORAL THREATS WITH HYBRID MACHINE LEARNING MODELS

Volume/Issue : 2025;02(7S)

Page No : 30-39

The rapid growth of online social networks has led to an explosion of user-generated content, with over 4.5 billion active users worldwide producing around 500 million tweets each day. This vast influx of data poses major challenges for traditional content moderation methods, which struggle to manage the volume and complexity of information. As a result, issues such as cyberbullying and misinformation persist. Conventional classification techniques often depend on manual efforts that are time-consuming and susceptible to human error, delaying effective responses. To address this, the study introduces a novel deep learning-based intelligent document classifier aimed at automatically categorizing social media content, with a focus on Twitter. The system employs advanced text preprocessing and N-gram feature extraction, utilizing a Convolutional Neural Network (CNN) to classify tweets into multiple categories, including religion, age, gender, ethnicity, and signs of cyberbullying. This automated solution improves the speed and accuracy of content moderation, contributing to more efficient and responsible online community management

Keywords : Cyberbullying Detection, Social Media Analysis, User-Generated Content, Text Classification

Author : Dr. B. Rama, Maheshuni Arun, Yellamelly Poorna Mani Tej, Chintham Sai Manish

Title : DEEPSCAN: AN INTELLIGENT DOCUMENT CLASSIFIER FOR CYBERBULLYING DETECTION IN SOCIAL MEDIA

Volume/Issue : 2025;02(7S)

Page No : 40-51

The global delivery industry is grappling with severe supply chain inefficiencies, contributing to the annual waste of over 1.3 billion tons of food—approximately 30% of total global production—leading to economic losses of around $940 billion. A significant portion of this issue stems from the fact that 40–50% of inventory-related decisions still rely on manual forecasting, resulting in frequent overstocking or understocking, increased spoilage, and logistical disruptions. Traditional forecasting methods are often static and incapable of adapting to dynamic environmental and market fluctuations, making them unsuitable for real-time driver demand prediction. To address these shortcomings, this research introduces a robust regression-based time series forecasting framework designed to optimize delivery supply chain operations. Utilizing a publicly available driver demand dataset that integrates temporal sales data, weather conditions, and promotional events, the data is preprocessed thoroughly through normalization, imputation of missing values, creation of lag-based features, and outlier handling to ensure high-quality model inputs. Initially, the Light Gradient Boosting Machine (LGBM) Regressor serves as a baseline model due to its strong predictive performance. However, to enhance forecasting precision, a Nonlinear Autoregressive model with Exogenous Inputs (NARX) Regressor is proposed. This model incorporates both historical internal demand data and external variables such as holidays and weather conditions, enabling dynamic and multivariate forecasting. NARX is particularly effective at capturing nonlinear patterns and supports recursive forecasting across multiple time steps. Comparative performance analysis shows that while the LGBM Regressor performs well (MSE: 7.06e-05, MAE: 0.00449, RMSE: 0.00840, R²: 0.948), the NARX Regressor significantly improves prediction accuracy, achieving MSE: 6.32e-05, MAE: 0.00350, RMSE: 0.00795, and an R² score exceeding 1.00. These results highlight the superior generalization and nearperfect predictive capabilities of the NARX model, demonstrating its potential to substantially reduce forecasting errors, minimize food waste, and enhance decision-making in supply chain management

Keywords : Driver Demand Forecasting, Food Delivery Platforms, Supply Chain Optimization, Nonlinear Autoregressive Models, Time Series Analysis

Author : G. Divya, Chennoju Bhanu Teja, Ryagati Rakesh, Bhukya Ruthwik

Title : SMART DISPATCH: ENHANCING DRIVER DEMAND FORECASTING IN RIDESHARING AND DELIVERY NETWORKS

Volume/Issue : 2025;02(7S)

Page No : 52-61

Sentiment analysis plays a crucial role in understanding public opinion on social media platforms like Twitter, where users express views in multiple languages that reflect global market dynamics. Realtime sentiment classification across diverse languages enables businesses and policymakers to extract actionable insights, highlighting the need for robust systems capable of handling linguistic diversity and evolving temporal patterns. A key challenge lies in accurately identifying sentiment from multilingual tweets in real time, accounting for variations in language (e.g., English, Spanish, French) and cultural differences in sentiment expression, while maintaining scalability and precision. Traditional sentiment analysis methods often rely on language-specific, rule-based, or lexicon-driven approaches using predefined sentiment dictionaries, usually paired with simple classifiers like Naive Bayes. These methods are inefficient for multilingual datasets, requiring separate models per language, and they struggle with context, sequence, imbalanced data, and cross-linguistic sentiment interpretation, resulting in inconsistent performance in global applications. The proposed system, is a Python-based application with a Tkinter GUI that processes multilingual tweets. It uses NLTK for preprocessing (including case normalization, stopword removal, and stemming), TF-IDF for feature extraction, and classification models such as Decision Tree, Random Forest, and Multilayer Perceptron (MLP). Despite the title, RNN has not been implemented. The system works with a small dataset of 17 tweets in English, Spanish, French, German, and Italian, each labeled with a sentiment score from 1 to 5 stars. The MLP classifier achieved an accuracy of 87.89%, demonstrating the potential of the system for multilingual sentiment monitoring. While this tool marks progress in multilingual sentiment analysis and offers a user-friendly interface for real-time global insights, its effectiveness is limited by the small dataset and the absence of RNN architecture. Future improvements involving larger datasets and RNN integration could significantly enhance its performance and real-time capabilities for global market applications

Keywords : Multilingual Sentiment Analysis, Real-Time Sentiment Classification, Natural Language Processing, Recurrent Neural Networks

Author : U. Meena, Sandamoni Adarsh, Ambati Gaurav, Karne Rathish

Title : NEUROLINGO: MULTILINGUAL SENTIMENT ANALYSIS USING RECURRENT NEURAL NETWORKS

Volume/Issue : 2025;02(7S)

Page No : 62-68

According to recent statistics, over 4.9 billion people actively use social media globally, generating nearly 500 million tweets and 4 billion Facebook interactions daily. Despite this immense data flow, only about 20% of social media trend predictions achieve reliable accuracy due to dynamic content patterns and informal language structures. Existing models struggle with the high variability, short-lived nature of trends, and noise in unstructured text data, often lacking scalability and contextual understanding. This research introduces a deep learning-based Social Media Trend Analyzer designed to accurately predict trend emergence over time (target: Trend – Yes or No) by overcoming these limitations. The system utilizes a curated dataset sourced from Twitter and Reddit, labeled with trending status based on hashtag propagation and engagement rates. Natural Language Processing (NLP) techniques such as lowercasing, stop-word removal, lemmatization, and punctuation filtering are applied during preprocessing. Text inputs are vectorized using a Tokenizer with word embeddings and padding sequences to ensure uniform input length. The baseline model employs XGBoost to classify trends, but its performance is limited by its inability to capture temporal semantics. To address this, we propose a hybrid RNN-LSTM model for deep temporal feature extraction, integrated with a Random Forest Classifier (RFC) for robust decisionmaking. The RNN-LSTM captures contextual word dependencies over time, which are then classified by RFC for final trend prediction. Experimental results demonstrate a significant improvement in accuracy, F1-score, and recall compared to traditional methods, confirming the efficacy of our approach in trend forecasting using unstructured social media data.

Keywords : Trend Prediction, Social Media Analytics, Time-Series Forecasting, Attention Mechanism, Virality Prediction

Author : G. Sudheer Kumar, Nithin Reddy Y, Ravikiran Mudhiraj G, Sai Prasad Ch

Title : DEEP LEARNING-BASED SOCIAL MEDIA TREND ANALYZER TO PREDICT THE TRENDS OVER TIME

Volume/Issue : 2025;02(7S)

Page No : 69-79

The global telemedicine market is expected to reach $559.52 billion by 2027, with healthcare costs rising at an estimated rate of 5.5% annually. As telemedicine becomes an integral part of modern healthcare, accurately forecasting future costs is crucial for optimizing healthcare expenditures and resource allocation. Existing cost estimation models often lack precision, failing to account for the dynamic nature of healthcare expenses and the conditions under which telemedicine provides a costeffective alternative. This study introduces a deep learning-based approach for estimating future healthcare costs associated with telemedicine services. The model employs a regression-based cost prediction framework while simultaneously classifying scenarios where telemedicine is a viable solution. The dataset undergoes comprehensive preprocessing, including data normalization, handling of missing values, and feature engineering, to improve model robustness and predictive accuracy. Advanced machine learning techniques, such as Ridge Regressor and Convolution Neural Networks (CNNs), are leveraged to capture complex patterns in healthcare expenditures. The proposed system offers a data-driven decision-making framework that enables healthcare providers, policymakers, and insurance companies to evaluate the financial feasibility of telemedicine solutions. By defining cost thresholds, the model assists in determining when telemedicine services should be prioritized over traditional in-person care, thereby reducing unnecessary expenses and enhancing healthcare accessibility. By integrating deep learning into cost estimation, this study contributes to a more efficient and predictive healthcare system, ensuring sustainable telemedicine adoption. Future research will focus on refining model interpretability, incorporating real-time data, and expanding datasets to enhance the model’s adaptability to various healthcare settings. This approach ultimately aims to drive strategic decision-making for telemedicine deployment, benefiting both healthcare providers and patients in a rapidly evolving digital healthcare landscape.

Keywords : Telemedicine, Healthcare Cost Prediction, Consultation Cost Estimation, Healthcare Economics, Digital Health Services, AI in Telemedicine

Author : A. Hareesha, Ganesh Ch, Sai Kumar Voore Medari, Ajay P

Title : TELECAREECONOMICS: AI-POWERED ESTIMATION OF PERSONALIZED FUTURE HEALTHCARE COSTS IN REMOTE MEDICAL SERVICES

Volume/Issue : 2025;02(7S)

Page No : 80-87

Autonomous vehicle navigation has emerged as a critical area of research, with global investments surpassing $80 billion in self-driving technologies. Studies indicate that over 90% of road accidents are caused by human error, emphasizing the need for automation. Additionally, navigation systems with advanced path-finding can reduce travel time by 15–25% in urban environments. However, existing manual analysis methods such as human visual monitoring, rule-based navigation, and static paper-based mapping suffer from subjectivity, lack of adaptability, and inefficiency in dynamic or unstructured environments. These techniques are time-consuming, prone to human error, and incapable of scaling for real-time autonomous decision-making. To address these limitations, the proposed system utilizes a structured preprocessing pipeline involving data cleaning, label encoding, and class distribution visualization. The input data, tailored to the application, undergoes null value imputation and label transformation using a label encoder to convert categorical direction labels into numerical form. A count plot is generated to assess class balance, providing insight into the need for SMOTE data augmentation. The final dataset is split into features and labels, ready for training with existing KNN and proposed MLP classifiers. This preprocessing stage ensures clean, balanced, and meaningful input that significantly enhances classification accuracy and real-time applicability in vehicle path-finding systems

Keywords : Autonomous Vehicle Navigation, Self-Driving Cars, Deep Reinforcement Learning, Edge AI for Vehicles, Trajectory Prediction

Author : Dr. S. Sankar Ganesh, Shiva Kumar B, Shivanand G, Vamshi S

Title : AUTOPATHAI: AN INTELLIGENT NAVIGATION FRAMEWORK FOR AUTONOMOUS VEHICLES USING AI-DRIVEN PATH OPTIMIZATION

Volume/Issue : 2025;02(7S)

Page No : 88-97

In today’s dynamic job market, finding the right fit between candidates and job roles has become a significant challenge for organizations. The increasing number of applicants for each job opening, coupled with diverse candidate profiles, makes it difficult for human resource teams to efficiently evaluate and select the most suitable individuals. Traditional recruitment processes primarily rely on manual screening and keyword-based filtering, which often fail to consider multiple important aspects such as skills alignment, experience, education, certifications, and availability. These limitations result in time-consuming hiring cycles, inconsistent selections, and potential mismatches between job requirements and candidate capabilities.The core problem lies in the inefficiency and subjectivity of traditional hiring methods. Manual evaluation is not scalable and is prone to human bias. Moreover, conventional systems lack the ability to analyze complex relationships among different candidate features and job expectations. As a result, there is a growing need for intelligent, data-driven systems that can automate and optimize the hiring process.To address these challenges, this research proposes an AIpowered job matching system that leverages supervised machine learning models to predict a Job Fit Score for each applicant. The system incorporates key candidate attributes such as skills score, experience, education level, certification strength, job location match, and immediate availability. By analyzing these inputs, the system predicts how well a candidate matches a specific job profile. Three different ML models are employed—Gradient Boosting Regressor (GBR), K-Nearest Neighbors (KNN), and Multi-Layer Perceptron (MLP)—with the MLP model achieving the best overall performance in terms of prediction accuracy and error metrics.This intelligent solution reduces manual effort, improves matching accuracy, and ensures fairer and more objective hiring decisions. It enhances HR efficiency by prioritizing candidates with the highest fit scores and offers a scalable model that can be adapted across industries. The proposed system signifies a major step toward transforming the traditional recruitment process through AI-driven insights and predictive analytics, enabling organizations to build more effective and future-ready teams

Keywords : Job-candidate classification, Resume parsing, Skill-to-role mapping, Predictive hiring model, Workforce analytics, Candidate-job fit prediction

Author : G. Sujatha, Bhanu Teja Gajawada, Maneendra P, Sai Sidhartha S

Title : JOBFITAI: AUTOMATED SKILL-BASED CLASSIFICATION AND MATCHING OF CANDIDATES TO RELEVANT JOB OPENINGS

Volume/Issue : 2025;02(7S)

Page No : 98-107

In today's digital marketing world, over 80% of online users prefer personalized ads, and personalized campaigns drive up to 20% higher conversion rates than generic ones. Traditional ad targeting relied on manual observations—such as reviewing customer records, analyzing past sales, or measuring viewership statistics. These methods were time-consuming, lacked real-time adaptability, and often led to poor campaign performance due to delays and human errors.Earlier machine learning approaches, like the Gradient Boosting Classifier, improved prediction accuracy over manual methods but struggled with overfitting, slow training on large datasets, and lack of adaptability to rapidly changing user preferences. While moderately accurate, they often failed in real-time deployment and had limited personalization at scale.Our proposed system uses a hybrid model combining a Feedforward Neural Network (FNN) with an Extra Trees Classifier to enhance both ad targeting accuracy and delivery efficiency. The FNN captures non-linear relationships in user behavior and preferences, while the Extra Trees Classifier ensures robust feature selection and high-speed predictions. This model uses data such as browsing history, click-through rate, demographics, and previous ad interactions to generate a personalized ad recommendation. Compared to traditional models, our system achieved a classification accuracy of 91%, a 28% increase over Gradient Boosting, and a significant reduction in training time. This personalized ML pipeline optimizes ad delivery while adapting dynamically to user trends, improving both customer engagement and return on ad spend.

Keywords : Personalized Ad Targeting, User Behavior Analysis, Engagement Prediction, Recommendation Systems, Targeted Advertising

Author : Dr. Pulime Satyanarayana, Yerram Srujankumar, Nivarthi Suryachandra Kumar

Title : SMARTADTARGET: AN INTELLIGENT ML FRAMEWORK FOR REAL-TIME PERSONALIZED ADVERTISEMENT OPTIMIZATION

Volume/Issue : 2025;02(7S)

Page No : 108-118

This thesis presents the development of a Virtual Reality (VR) User Experience Classification system employing deep learning techniques aimed at enhancing VR design quality. The system utilizes a comprehensive VR user experience dataset, which comprises over 10,000 user interaction records collected from diverse VR environments. The dataset exhibits a notable class imbalance, with approximately 60% of samples belonging to the majority class and 40% representing minority classes. Traditional classification methods often struggle to effectively manage such imbalanced data, resulting in suboptimal accuracy, precision, recall, and F1 scores below 75%, thereby limiting their practical applicability in real-world VR design evaluation. Existing manual and conventional approaches for VR user experience classification face significant challenges, including inadequate handling of imbalanced datasets, poor generalization across varying user behaviors, and limited ability to extract complex feature patterns inherent in VR interactions. These issues often lead to inaccurate user experience predictions, hindering the iterative design process crucial for improving VR applications. To overcome these limitations, this thesis proposes a novel hybrid approach combining Synthetic Minority Over-sampling Technique (SMOTE) with a Deep Neural Network (DNN) classifier. SMOTE addresses the class imbalance by generating synthetic samples for underrepresented classes, effectively balancing the dataset without information loss. The DNN model then automatically extracts hierarchical and meaningful features from the balanced data, enhancing the classifier’s ability to recognize subtle patterns and complex relationships within the VR user experience data. The proposed SMOTE with DNN framework was rigorously evaluated through traintest splitting methodology, and its performance was compared with existing baseline models. Experimental results demonstrate a significant improvement, with the proposed method achieving over 99% accuracy, precision, recall, and F1 score. These outcomes validate the effectiveness of the hybrid approach in accurately classifying VR user experiences, which in turn can provide actionable insights for design enhancement and user satisfaction improvement

Keywords : Virtual Reality (VR), User Experience (UX), Immersive Technology, Emotion Recognition, Neural Networks

Author : Dr. B. Rama, Mangisetty Santosh, Degalvadi Vinod Kumar

Title : NEUROVR: DEEP LEARNING-DRIVEN INSIGHTS FOR ENHANCING VIRTUAL REALITY DESIGN THROUGH USER EXPERIENCE CLASSIFICATION

Volume/Issue : 2025;02(7S)

Page No : 119-128

Liver disease poses a growing challenge to global healthcare systems, with millions affected annually—particularly in India, where it accounts for over a million cases each year. early and accurate diagnosis is essential for effective treatment, yet traditional diagnostic techniques such as blood tests, ultrasound imaging, and biopsies often suffer from limitations like time delays, human error, and dependency on manual interpretation. To address these challenges, this research explores the application of artificial intelligence (AI) in liver disease detection using advanced pre-trained deep learning models such as VGG16, ResNet60, and GoogLeNet. these models, widely recognized for their success in medical image analysis, are further enhanced through optimization using the Social Spider Optimization (SSO) algorithm. the system is trained and evaluated on liver scans from the NIfTI dataset, enabling automated multi-class classification of liver abnormalities. The proposed approach demonstrates significant improvements in classification performance, with VGG16 optimized via SSO achieving the highest diagnostic accuracy, precision, and recall among the tested models. this hybrid framework offers a powerful tool for assisting medical professionals in making faster, more reliable diagnostic decisions, contributing to improved outcomes in liver disease management

Keywords : Medical Image Analysis, Dimensionality Reduction, Liver Disease Prediction, Convolutional Neural Network (CNN).

Author : Krunal Kanubhai Prajapati,Kamalesh V. N,Shankar Nayak Bhukya

Title : HYBRID SOCIAL SPIDER OPTIMIZATION AND PRETRAINED CNNS FOR ACCURATE LIVER ABNORMALITY DETECTION

Volume/Issue : 2025;02(7S)

Page No : 129-140

The Indian banking sector plays a crucial role in the country's economic growth, with stocks of major banks being actively traded on the National Stock Exchange (NSE). This study aims to analyze the equity performance of banking stocks listed on the NSE, focusing on their financial stability, growth potential, and investment attractiveness. The research is conducted at Northeast Broking Services Ltd, leveraging fundamental and technical analysis methods to evaluate stock performance. The study examines key financial indicators such as Price-to-Earnings (P/E) Ratio, Earnings Per Share (EPS), Return on Equity (ROE), and Net Interest Margin (NIM) to assess banks' financial health. Additionally, technical indicators like Moving Averages, Relative Strength Index (RSI), and Bollinger Bands are used to identify trends and market movements. The research also considers macroeconomic factors, including RBI policies, interest rate changes, and global economic trends, influencing banking stocks.

Keywords :

Author : GUNJA RAVI TEJA, R HARISH CHANDRA, DR. K. VEERAIAH

Title : A STUDY ON EQUITY ANALYSIS OF INDIAN BANKING SECTOR STOCKS IN NSE" AT NORTHEAST BROKING SERVICES LTD

Volume/Issue : 2025;02(7S)

Page No : 141-146

India’s e-commerce sector has experienced rapid growth, with the number of online shoppers projected to reach 175 million by 2020, driven by increased adoption of digital payments and mobile transactions. However, this growth has also led to a surge in online fraud, including phishing and card testing schemes. In 2018 alone, ad fraud losses amounted to $1.63 billion, with e-commerce accounting for 51% of that figure. Traditionally, fraud detection relies on manual systems where human analysts examine transactions based on predefined rules, such as repeated declines, unusual IP addresses, or mismatched billing and shipping details. These systems often use CVV verification and CAPTCHA tests, but are prone to human error, overlook complex fraud patterns, and struggle to scale with high transaction volumes. This research addresses these limitations by introducing an automated fraud detection system using machine learning. The proposed system classifies transactions as “fraudulent” or “not fraudulent.” Various models are explored: Support Vector Machine (SVM) identifies class boundaries using a hyperplane and performs well on small datasets but is sensitive to class imbalance; Logistic Regression provides interpretable probabilistic outputs but lacks capacity for complex patterns; Neural Networks can capture intricate relationships through multiple layers, though they are computationally intensive; and Gradient Boosting Classifier, paired with SMOTE for handling imbalanced data, constructs sequential trees and effectively models non-linear patterns

Keywords : E-commerce Fraud Detection, Imbalanced Data Handling, Digital Payment Security, Financial Transaction Monitoring

Author : K. Vamshee Krishna, Kishore Naik, Dinesh Kumar. K, Laxman Reddy. V

Title : GRADIENT BOOSTING WITH SMOTE FOR REAL-TIME FRAUD DETECTION IN E-COMMERCE

Volume/Issue : 2025;02(7S)

Page No : 147-157

With the rapid escalation of cyber threats, security breaches have increased by 67% over the past five years, and global cybercrime costs are projected to reach $10.5 trillion annually by 2025. Traditional manual threat detection systems, which depend on rule-based logic and human oversight, are often error-prone, slow to respond, and inadequate for processing large-scale security data. To overcome these limitations, we present a machine learning-driven behavioral analysis method for detecting security threats, integrating deep learning techniques to enhance detection accuracy. The proposed approach involves preprocessing the Security Threat Dataset—comprising four attack categories: Denial of Service (DoS), Probe, Remote-to-Local (R2L), and User-to-Root (U2R)—to clean and normalize the data. To address class imbalance, we implement ADASYN (Adaptive Synthetic Sampling), which synthetically augments minority class samples for a balanced dataset. The data is partitioned into training and testing sets to assess model performance. We benchmark conventional models like Naïve Bayes and Support Vector Machine (SVM) against our Convolutional Neural Network integrated with Generative Adversarial Network (CNN-GAN). This hybrid model effectively captures intricate patterns and hierarchical structures in network traffic. Experimental evaluation confirms that the CNN-GAN model surpasses traditional methods in accuracy, precision, and recall, establishing it as a powerful framework for real-time threat detection

Keywords : Cyber Threat Detection, ADASYN (Adaptive Synthetic Sampling), Intrusion Detection System (IDS), Supervised Pretrained Classifier (SPC).

Author : Ritesh Kumar, Kalyan Somagani, Kishan Nayak V, Madhusudhan Rao. A

Title : ADASYN-ENHANCED CNN-GAN FRAMEWORK FOR IMBALANCED INTRUSION DETECTION IN IOT NETWORKS

Volume/Issue : 2025;02(7S)

Page No : 158-165

Traditional cheque clearance systems are plagued by issues such as processing delays, high operational costs, and limited transparency, leading to inefficiencies and increased risk of fraud. To overcome these limitations, this paper proposes a Blockchain-Based Cheque Clearance System Using Ethereum that aims to transform the banking sector by providing a secure, transparent, and efficient framework for cheque transactions between banks and users. The system utilizes Ethereum's decentralized blockchain architecture and smart contracts to automate the clearance process, enabling faster and tamper-resistant transactions. Historically, cheque processing involved manual and centralized operations vulnerable to errors and fraudulent activities, often requiring inter-bank coordination that introduced delays and inconsistencies. While various digital methods have attempted to modernize this process, they remained centralized and failed to establish trust and transparency. This research leverages the decentralized, immutable, and transparent nature of blockchain technology to address these long-standing challenges. The proposed system enhances security, ensures data integrity, and reduces human involvement by recording each transaction on the blockchain, enabling real-time verification and clearance via smart contracts. By eliminating the need for a central authority, the system becomes more resilient, reliable, and trustworthy. Overall, this approach aims to revolutionize cheque clearance by delivering a seamless, auditable, and secure solution for modern banking

Keywords : Blockchain in Banking, Decentralized Finance (DeFi), Fraud Prevention in Finance, Ethereum Smart Contracts, Cheque Clearance Automation

Author : K. Sunil Kumar, Architha Gangone, Divya Thatipamula, Akshitha Mutupuri,Ganesh Kapuganti

Title : DECENTRALIZED CHEQUE SETTLEMENT VIA ETHEREUM SMART CONTRACTS: A SECURE BLOCKCHAIN FRAMEWORK

Volume/Issue : 2025;02(7S)

Page No : 166-173

The proliferation of Internet of Things (IoT) networks has transformed both industrial and domestic environments, but it has also introduced significant security challenges, particularly in defending against Distributed Denial of Service (DDoS) attacks. IoT sensor nodes, often lightweight and resource-constrained, are highly susceptible to volumetric and protocol-based intrusions. Traditional security approaches relying on static rules, signature-based detection, and manual threshold configurations—struggle to detect zero-day exploits and obfuscated attack patterns. These methods are typically ineffective at capturing subtle anomalies, lack scalability to process vast IoT traffic, and suffer from high false alarm rates and delayed responsiveness. To address these limitations, this work proposes an intelligent, automated, and adaptive DDoS detection framework powered by supervised machine learning. The system processes raw network traffic through a pipeline that includes label encoding, normalization, and dimensionality reduction using Principal Component Analysis (PCA) to optimize the quality and efficiency of training data. Six machine learning classifiers Naive Bayes, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), XGBoost, and AdaBoost are implemented and evaluated to identify the most effective model for real-time detection. A user-friendly interface developed using Tkinter allows for seamless dataset handling, model training, and performance visualization. The proposed framework not only enhances detection accuracy and reduces false positives but also supports rapid deployment and is suitable for integration with resource-limited IoT devices. By offering comparative model evaluation, the system equips security professionals with the tools to select the optimal algorithm for their specific network environments, significantly bolstering IoT resilience against evolving DDoS threats while ensuring scalability, adaptability, and operational efficiency

Keywords : DDoS Detection, Wireless Sensor Networks (WSN), Random Forest Classifier, Anomaly-Based Detection.

Author : Ch. Jyothi, Ganesh Reddy. D, Farhan Mohammad, Imran Bin Javed Md

Title : ENSEMBLE-ENHANCED MACHINE LEARNING FOR REAL-TIME DDOS DETECTION IN RESOURCE-CONSTRAINED WSNS

Volume/Issue : 2025;02(7S)

Page No : 174-181

Bot-driven cyberbullying has become a growing concern in digital communication, with statistics indicating that over 59% of teenagers have encountered online harassment—most commonly on social media platforms, accounting for over 70% of reported cases. Multi-class cyberbullying datasets typically contain up to 25,000 labeled entries across categories such as insults, threats, racism, and sexism, often suffering from severe class imbalance and linguistic diversity. Manual detection techniques are plagued by inconsistencies, subjectivity, and limited scalability amidst the rapidly growing volume of user-generated content. Traditional machine learning models struggle with shallow feature representations, poor performance on minority classes, and difficulty in detecting nuanced or implicit abuse. Additionally, current literature often overlooks the integration of deep ensemble learning with optimized preprocessing and contextual analysis. To overcome these challenges, this study introduces a Hybrid Multiclass Unmasking Bot Classification framework. The proposed approach combines feature-enriched N-gram extraction with a dual deep learning architecture incorporating both Deep Neural Networks (DNN) and Convolutional Neural Networks (CNN). The pipeline begins with comprehensive dataset ingestion and Exploratory Data Analysis (EDA) to understand class distribution and data imbalance. Text preprocessing, including tokenization, lemmatization, and noise elimination, is followed by vectorization using TF-IDF with bi-gram support, enabling the capture of both isolated and contextual semantics. The DNN component captures abstract semantic relationships, while the CNN detects local linguistic cues. This dual-stream structure promotes robust learning across varying categories of cyberbullying bots. Performance evaluation demonstrates that the proposed system significantly outperforms conventional classifiers in terms of precision, recall, and F1-score across all classes

Keywords : Cyberbullying Detection, Bot Classification, Deep Neural Network (DNN), Convolutional Neural Network (CNN), TF-IDF, Multi-class Classification, Natural Language Processing (NLP), Ensemble Deep Learning, Feature Extraction, Text Preprocessing

Author : K. Manohar Rao, Hruthik Sheelam, Aravind Seelam, Ajay Erumandla

Title : DETECTING CYBERBULLYING BOTS USING HYBRID CNN AND ENHANCED TEXT FEATURES

Volume/Issue : 2025;02(7S)

Page No : 182-189

This thesis introduces a Virtual Reality (VR) User Experience Classification framework powered by deep learning, aimed at improving the design quality and user satisfaction of VR systems. The approach utilizes a rich dataset consisting of over 10,000 user interaction instances sourced from various VR environments. A key challenge identified in the dataset is class imbalance, where nearly 60% of the data belongs to the majority class, and the remaining 40% is distributed among minority classes. To address this, the study proposes a hybrid method that integrates the Synthetic Minority Over-sampling Technique (SMOTE) with a Deep Neural Network (DNN) classifier. SMOTE is employed to synthetically generate new instances for the minority classes, thereby balancing the dataset without compromising original data integrity. Subsequently, the DNN model learns hierarchical and abstract features from the balanced dataset, enabling it to capture complex behavioral patterns and interactions effectively. The proposed SMOTE-DNN model was evaluated using a traintest split approach and benchmarked against traditional baseline models. The experimental outcomes show a significant enhancement in performance, achieving over 99% across accuracy, precision, recall, and F1 score. These results affirm the efficacy of the proposed hybrid model in classifying VR user experiences and offer valuable insights for future improvements in VR system design and personalization

Keywords : Virtual Reality (VR), User Experience, Deep Neural Network (DNN), Synthetic Minority Oversampling Technique (SMOTE), Classification, Imbalanced Dataset, User Behavior Analysis, Machine Learning, Human-Computer Interaction, Immersive Technology

Author : Dr. C. Bagath Basha, Bhagavanthu Goulla, Akash Neelagiri, Charan Menga

Title : DEEP LEARNING-BASED USER BEHAVIOR CLASSIFICATION TO IMPROVE VIRTUAL REALITY INTERACTION

Volume/Issue : 2025;02(7S)

Page No : 190-198

Injection attacks account for over 35% of all API-related vulnerabilities, with SQL injection alone contributing to 25% of critical exploits in web applications. According to OWASP, injection flaws have ranked in the Top 10 web application security risks for over a decade, costing organizations millions annually in remediation and data loss. Traditional detection systems still miss 12–18% of obfuscated or zero-day injection payloads due to static rule sets. Manual or rule-based detection systems often rely on signature matching or regex filters, which fail to adapt to evolving attack patterns. These methods are typically unable to detect contextual anomalies in API request sequences and struggle with false positives that burden security teams. Furthermore, they lack the temporal awareness needed to analyse sequential API interactions effectively. In this research, we propose a deep learning-based model leveraging Bidirectional Recurrent Neural Networks (Bi-RNNs) to detect injection attacks in API traffic. The (Bi-RNNs) are uniquely suited for this task as they process API input sequences in both forward and backward directions, capturing comprehensive contextual dependencies that unidirectional models often miss. The dataset comprises 500,000 API requests, equally distributed between normal and various types of injection attacks, including SQLi, XSS, Command Injection, and XMLi. The model is trained to extract semantic and syntactic patterns from raw API parameters, achieving a detection accuracy of 95.7%, with precision and recall values exceeding 95%. Unlike traditional intrusion detection systems, our model adapts dynamically to emerging attack vectors through retraining on new data and significantly reduces false positives. Given the exponential rise in API integrations across industries and the 22% year-over-year increase in injection-based API attacks, intelligent automated detection is no longer optional. Our findings address the urgent need for context-aware, sequence-sensitive security solutions in modern API ecosystems

Keywords : Injection Attacks, API Security, Bidirectional Recurrent Neural Networks (BRNNs), Detection Accuracy, Automated Threat Detection

Author : C. Vijayaraj, B. Sai Akshara, B. Prasad, Chiluka Rahul

Title : BIDIRECTIONAL RNN-BASED DETECTION OF INJECTION ATTACKS IN API TRAFFIC WITH CONTEXTUAL SEQUENCE ANALYSIS

Volume/Issue : 2025;02(7S)

Page No : 199-209

The global video-on-demand market is projected to surpass $257 billion by 2027, with platforms like Netflix attributing over 80% of content watched to recommendation engines. This highlights the growing demand for intelligent, accurate, and personalized movie recommendation systems. However, existing systems face persistent issues such as data sparsity, cold-start problems, and lack of content diversity, which hinder their effectiveness and scalability. To address these limitations, this research introduces a novel hybrid recommendation framework named Content-Based Collaborative Filtering integrated with Convolutional Neural Networks (CBCF-CNN). The proposed model innovatively combines the strengths of content-based filtering, collaborative filtering, and deep learning to deliver a robust and scalable recommendation system. CBCF-CNN leverages user-item interaction matrices along with highdimensional feature extraction from movie metadata and visual content using CNNs. This deep integration allows the system to capture latent patterns and semantic similarities between items, thus mitigating the cold-start issue and enriching the recommendation space. Furthermore, the architecture enhances personalization by learning user-specific behavior from sparse data and enables real-time inference through parallelized CNN processing. Unlike traditional models that treat collaborative and content-based filtering separately, CBCF-CNN creates a unified representation that dynamically adapts to user preferences while promoting recommendation diversity and relevance. The model's ability to scale with massive datasets while maintaining low latency and high accuracy makes it suitable for deployment in large-scale platforms, offering a significant advancement over existing recommendation techniques

Keywords : Movie Recommendation, CBCF-CNN, Content-Based Filtering, Collaborative Filtering, Convolutional Neural Network, Cold-Start Problem, Data Sparsity, Personalization, Deep Learning, Word Cloud, User-Item Interaction, Recommendation Diversity.

Author : K. Vamshee Krishna, Rishendra Varshith Raju.E, Sai Sri Prabath. Y, Singarweni Rahul

Title : ENHANCING MOVIE RECOMMENDATIONS USING CBCF-CNN WITH WORD CLOUD VISUALIZATIONS

Volume/Issue : 2025;02(7S)

Page No : 210-222

With the rapid growth of cloud computing, over 90% of enterprises are shifting their workloads to cloud service providers (CSPs), driven by cost-efficiency and scalability. However, over 40% of cloud users report missed cost-saving opportunities due to static VM placement and reluctance to migrate despite better offers. The primary challenge lies in the high cost of inter-cloud VM migration, which deters customers from switching to lower-cost providers, thereby reducing potential profits. Additionally, lack of automated migration frameworks and real-time decision-making tools further exacerbates inefficiencies. To address these issues, this paper introduces a Federated Cloud VM Migration framework designed to maximize user profit by dynamically migrating virtual machines (VMs) from high-cost data centers (DCs) to low-cost ones within a federated cloud environment. The proposed system evaluates DCs based on cost and region (distance), identifies optimal source and destination pairs, and performs migrations only if they lead to overall cost savings, factoring in migration time, data size, and cost per hour. Unlike traditional systems, our approach considers costdistance trade-offs and implements a priority-based scheduling algorithm to ensure seamless migration even when storage constraints exist at the destination. The implementation includes interactive modules for user sign-up/login, CP/DC generation with cost-distance mapping, file uploads, and live VM migration tracking. This novel federated cloud strategy enables adaptive, costaware, and space-efficient VM migrations, thus significantly enhancing profit margins for users and increasing cloud resource utilization across providers

Keywords : Adaptive federated cloud, VM migration, cost-distance optimization, priority scheduling, inter-cloud migration, cost-efficient framework, cloud resource utilization, dynamic VM placement, migration tracking, federated cloud strategy

Author : K Vijaya Bhaskar Reddy, Madhireddy Pavan Kumar Reddy, Kota Prabhas Reddy, Pratap Kumar

Title : ADAPTIVE FEDERATED CLOUD FRAMEWORK FOR COSTEFFICIENT VM MIGRATION USING COST-DISTANCE AND PRIORITY SCHEDULING

Volume/Issue : 2025;02(7S)

Page No : 223-234

The liver plays a crucial role in human health, yet liver diseases such as fatty liver, cirrhosis, and hepatitis are becoming increasingly prevalent worldwide. Factors such as irregular diets, excessive alcohol consumption, and unhealthy lifestyles contribute significantly to liver dysfunction. The high mortality rate associated with liver diseases necessitates early detection for timely medical intervention. Common diagnostic methods, including biopsy, X-ray, CT scans, and MRI scans, assist physicians in identifying liver abnormalities at an early stage. This research explores an advanced liver disease detection framework leveraging artificial intelligence (AI) and pre-trained deep learning algorithms. CT scans of patients serve as the primary dataset for classification. The study employs the Social Spider Optimization (SSO) Algorithm to optimize and reduce the feature set from CT images, enhancing computational efficiency and classification performance. After feature extraction and reduction using SSA, three well-established Convolutional Neural Networks (CNN) are utilized for liver image classification. The CNN models are trained and fine-tuned using the optimized feature set obtained through SSA. A publicly available dataset of liver CT images is employed to evaluate the performance of these models based on various assessment metrics. The models are trained with a structured approach to analyze the impact of feature reduction. By integrating SSA for feature optimization with pre-trained CNN architectures, this study contributes to improving liver disease detection. The proposed methodology emphasizes the significance of AI-driven approaches in medical imaging and disease diagnosis

Keywords :

Author : C. Bagath Basha, Sai Surekha Pedagadi, Rakshitha Kandakatla, Poojitha Beerla

Title : DEEP LEARNING-BASED LIVER CT SCAN ANALYSIS WITH SOCIAL SPIDER OPTIMIZATION FOR FEATURE REDUCTION

Volume/Issue : 2025;02(7S)

Page No : 235-243