Journal of Science Engineering Technology and Management Science

The rapid expansion of online communication platforms and digital technologies has significantly increased cybersecurity-related discussions across social media, forums, and reporting platforms. These sources generate large volumes of unstructured textual data containing valuable insights into cyber threats, vulnerabilities, and attack activities. Traditional cybersecurity monitoring methods relied on manual analysis and rule-based filtering techniques, where analysts used predefined keywords and pattern matching. While useful for basic monitoring, these approaches are inefficient for handling large-scale data and fail to capture contextual meanings. A major challenge in cybersecurity intelligence is extracting relevant information from complex and noisy textual datasets. Conventional systems struggle with high data volume, inconsistent language, and irrelevant content, limiting their ability to accurately classify cyberattack categories. This highlights the need for intelligent, data-driven frameworks. To address these issues, this research proposes a machine learning-based framework for cyberattack detection using textual data. The system incorporates natural language preprocessing for data cleaning and normalization, followed by semantic feature extraction using Sentence-Bidirectional Encoder Representations from Transformers (SBERT) embeddings trained with an Open Question Answering objective. To overcome class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is applied. The framework evaluates multiple classifiers, including Stochastic Gradient Descent (SGD), Complement Naïve Bayes (CNB), and a hybrid model combining Dense Neural Networks (DNN) with Linear Discriminant Analysis (LDA). Experimental results show that the hybrid approach enhances classification performance by leveraging deep feature representations alongside statistical methods. The proposed system improves automated cybersecurity text analysis and supports efficient cyberattack detection in large-scale datasets

Keywords : Cybersecurity, Text Mining, Unstructured Data Analysis, Natural Language Processing, Cyber Threat Intelligence, Feature Extraction, Data Preprocessing, Class Imbalance.

Author : Bulusu Rama, Gunti Bhagya Laxmi, Dandegatla Disha, Balaraju Divisha Varma

Title : CyberSense-X: A Semantic-Enriched Ensemble Learning Framework for Real-Time Detection of Malicious Twitter Activity

Volume/Issue : 2026;03(04)

Page No : 924-935

The rapid growth of e-commerce and online delivery platforms has increased the importance of efficient last-mile courier services. Companies such as Amazon and Flipkart depend on courier performance to ensure timely delivery and maintain customer satisfaction. The surge in online orders generates large volumes of structured delivery data and unstructured customer feedback, containing valuable insights into courier efficiency, delivery accuracy, service reliability, and overall experience. Traditional evaluation methods, including manual review and basic statistical summaries like average ratings, fail to effectively analyze large datasets or interpret complex textual feedback. To address this challenge, this study proposes a machine learning (ML)-based analytical framework for sentimentdriven performance benchmarking of last-mile couriers. The framework integrates data preprocessing, text feature extraction, and multiple classification models to analyze courier-related data. A graphical user interface (GUI) allows users and administrators to interact with the system, while Redis serves as a lightweight in-memory database for secure authentication and credential management. The study implements several ML algorithms, including Gaussian Naive Bayes (GNB), Random Forest (RF), and K-Nearest Neighbors (KNN), as baseline classifiers. Additionally, a hybrid model, Leaf2VecKNN, is proposed to enhance classification performance. In this approach, RF converts courier features into leaf-node representations, which are encoded and classified using KNN similarity learning. Experimental results show that the proposed Leaf2VecKNN model achieves the highest accuracy of 91.21%, outperforming baseline models. The framework provides an efficient solution for automated courier feedback classification and supports data-driven decision-making to improve delivery service quality

Keywords : Last-mile delivery, courier performance, e-commerce logistics, sentiment analysis, customer feedback analysis, delivery efficiency, service reliability, text preprocessing, feature extraction

Author : G. Divya, Vakiti Kulashekar Reddy, Dusa Maniteja, Thaduka Nuthan Kumar

Title : Hybrid Structured–Unstructured Delivery Feedback Analysis for ECommerce LMD Enhancement

Volume/Issue : 2026;03(04)

Page No : 936-946

The rapid growth of digital platforms has led to the generation of large volumes of employee feedback data, making workforce satisfaction analysis an important area of study. Traditionally, organizations relied on manual surveys and basic statistical methods to evaluate employee satisfaction, which were often time-consuming and limited in capturing complex textual insights. With advancements in Natural Language Processing (NLP) and Machine Learning (ML), automated analysis has become feasible. However, existing approaches struggle with unstructured text, class imbalance, and multidimensional prediction tasks. The primary problem addressed in this study is the accurate prediction of workforce satisfaction factors such as work-life balance, skill development, salary and benefits, job security, career growth, and overall satisfaction from textual employee reviews. Traditional systems fail to process large-scale data efficiently and lack consistency in predictive performance. This creates the need for an intelligent framework capable of handling complex textual patterns and multi-label classification. To overcome these challenges, the proposed system integrates NLP preprocessing, transformer-based feature extraction using Google PaLM (Pathways Language Model – PaLM), and SMOTE (Synthetic Minority Over-sampling Technique). Multiple ML models including Quadratic Discriminant Analysis (QDA), Linear Discriminant Analysis (LDA), and Histogram-Based Gradient Boosting (HGB) are implemented and compared with the proposed Transformer-Guided Adaptive Model (TGAM). The results show that traditional models achieve moderate accuracy ranging from approximately 51% to 56%, while the proposed TGAM model achieves 100.00% accuracy across all target columns including work-life balance, skill development, salary and benefits, job security, career growth, and work satisfaction. This significant improvement highlights the effectiveness of the proposed approach in handling complex workforce data. The system also includes evaluation metrics and visualization techniques for better interpretability.

Keywords : Employee Review Analysis, work-life balance, Natural Language Processing (NLP), SBERT (Sentence-BERT), Stochastic Gradient Descent (SGD), Job Security.

Author : A. Hareesha, Prasanna Laxmi Medichelmi, Mudireddy Sandeep Reddy, Rohini Moola

Title : Fine-Grained Opinion Mining from Employee Reviews for Organizational Decision Support

Volume/Issue : 2026;03(04)

Page No : 947-960

Gait is the locomotion attained through the movement of limbs, and gait analysis examines the patterns (normal/abnormal) depending on the gait cycle. It contributes to the development of various applications in the medical, security, sports, and fitness domains to improve the overall outcome. Among many available technologies, two emerging technologies that play a central role in modernday gait analysis are wearable sensors, which provide a convenient, efficient, and inexpensive way to collect data, and Machine Learning Methods (MLMs) which enable high-accuracy gait feature extraction for analysis. The dataset comprises continuous recordings from inertial measurement units (IMUs) capturing acceleration, angular velocity, and orientation across multiple participants. Existing approaches, including Adaptive Boosting (AdaBoost), K-Nearest Neighbor (KNN), Logistic Regression (LR), and Naive Bayes (NB) classifiers, serve as baseline models for gait detection. The proposed algorithm is Greedy Tree Classifier. The output is a multi-class gait activity label, including walking, running, climbing, and other activities. Integration into a Flask-based web application enables real-time streaming, preprocessing, and activity prediction for wearable devices. Experimental evaluation shows that the proposed approach outperforms existing classifiers in terms of accuracy, F1-score, and robustness across long-duration recordings. This framework supports continuous monitoring in healthcare, sports, and rehabilitation applications, enabling reliable activity tracking, early detection of mobility issues, and personalized interventions.

Keywords : Greedy Tree Classifier, Multi-Gait Activity Detection, Gait Analysis, Human Activity Detection.

Author : P. Babu, Ch. Nikhitha, D. Kavya, G. Haneesha, I. Lakshmi Sudha

Title : Intelligent Multi-Gait Identification using Greedy Tree Learning on Streaming Wearable Sensor Data

Volume/Issue : 2026;03(04)

Page No : 961-973

Online mobile technology reviews have grown rapidly, with millions of new reviews posted each month and more than 80 percent of consumers relying on them before making purchase decisions. The large volume and unstructured nature of these reviews create significant challenges for timely and consistent analysis, particularly for high-demand products such as iPhones. Automated analysis of mobile tech reviews is essential for application scenarios such as customer sentiment monitoring, product quality assessment, market trend analysis, and decision support for both consumers and manufacturers. Accurate interpretation of review titles and ratings can directly influence purchasing behavior and product improvement strategies. Traditional manual review analysis suffers from high time consumption, human bias, poor scalability, and inconsistency when handling large-scale datasets. These limitations make it difficult to extract reliable insights or identify hidden sentiment patterns from massive collections of mobile tech reviews. To address these challenges, this work proposes a Dual-Target Efficiently Learning an Encoder That Classifies Token Replacements Accurately (ELECTRA) prediction of iPhone mobile tech reviews. The approach leverages an NLP-based dataset of mobile tech reviews and employs the ELECTRA transformer model for contextual text representation. Multiple machine learning classifiers, including AdaBoost Classifier, Tree Alternative Optimization (TAO) Tree Classifier, and Extra Tree Classifier, are integrated to enhance predictive performance and enable comparative evaluation. The unified framework processes review text efficiently and generates two outputs within a single pipeline: classification of review titles and prediction of user ratings. The proposed system improves analytical accuracy, reduces manual effort, and provides scalable, reliable insights for mobile technology review analysis.

Keywords : mobile tech review analysis, iPhone review prediction, sentiment classification, Machine Learning, Transformer

Author : B. Poojitha, R. Siva Sankar, P. Hemanth, V. Sai Tharun, T. Ajay

Title : ReviewSenseNet: Dual-Task Transformer Learning for Title Classification and Rating Prediction

Volume/Issue : 2026;03(04)

Page No : 974-988

Road accidents account for over 1.19 million deaths globally every year, with delayed emergency response being a major contributing factor. Studies show that automated siren detection can reduce emergency vehicle response time by up to 25 percent in high traffic zones. Smart traffic management systems require real time emergency sound detection to control signals and clear congestion. Surveillance systems, ambulance routing, and disaster response networks rely on accurate siren recognition in noisy outdoor environments. Manual monitoring of audio signals is slow, inconsistent, and highly dependent on human attention. Background noise, overlapping sounds, and fatigue lead to frequent misclassification and delayed emergency response. The proposed system employs a publicly available urban sound dataset consisting of ambulance, fire truck, and traffic audio samples. Audio pre-processing is performed, followed by feature extraction using Mel-Frequency Cepstral Coefficients (MFCC) and Chromagram-based pitch representations (Chroma features) to obtain compact spectral representations. The dataset is divided into training and testing sets for evaluation. Existing classification models include Generalized Learning Vector Quantization (GLVQ), Perceptron, Multi-Layer Perceptron (MLP), and Deep Neural Network (DNN), with a Perceptronbased DNN executed as the primary baseline model. These models are evaluated under noisy urban conditions, demonstrating improved robustness and accuracy for real-time emergency sound detection.

Keywords : Smart traffic management, Real time emergency, Congestion control, Rapid urbanization Disaster response

Author : V. Gayatri, K. Dileep, Sk. Abdul Muqeem, S. Pavan Gowtham Kumar, Sk. Muheed

Title : URBAN-NET: A LOW LATENCY DEEP LEARNING FRAMEWORK FOR EMERGENCY SIGNAL EXTRACTION IN NOISY ENVIRONMENTS

Volume/Issue : 2026;03(04)

Page No : 989-999

The Internet of Things (IoT) has transformed into a vast and dynamic ecosystem in which billions of interconnected devices continuously exchange data to enable real-time monitoring, intelligent decisionmaking, and automated control. These devices operate through Local Area Networks (LAN), Internetbased communication, and Ad-hoc networks, generating massive volumes of sensor data that demand efficient processing to maintain system responsiveness. Conventional cloud-centric approaches face limitations such as high latency, bandwidth constraints, and increased energy consumption, making them unsuitable for time-sensitive applications including healthcare, autonomous systems, and industrial automation. To address these challenges, a hierarchical computing paradigm integrating Edge, Fog, and Cloud layers has been adopted. Edge computing ensures low-latency processing near data sources, while fog computing provides intermediate computational support for data aggregation and analysis, and the cloud enables large-scale analytics and long-term storage. However, optimal task scheduling across these layers remains a critical challenge. Existing approaches such as Passive Aggressive Classifier (PAC), Naive Bayes Classifier (NBC), and K-Nearest Neighbors (KNN) provide baseline solutions but lack adaptability in dynamic environments. To overcome these limitations, a hybrid Convolutional Recurrent Network with Greedy Rule Interpretable Machine (CRN-GRIM) is proposed, combining deep spatiotemporal learning with interpretable rule-based decision-making. This approach enhances task allocation efficiency by considering priority, resource demand, and network conditions, thereby improving system scalability, latency reduction, and overall performance in IoT-driven Edge–Fog ecosystems.

Keywords : Task scheduling. Edge-fog environments, Temporal learning, Local area networks, CRNGRIM, Hierarchical computing

Author : B. Lakshman Rao, P. Wajid, Sk. Umar Farooq, Sk. Vasi UR Rahman, Sk. Abubakar Siddiq, Yeneti Lokesh

Title : A Unified Deep Temporal Inference and Rule-Based Decision Framework for Resource-Aware Task Scheduling in Edge–Fog Networks

Volume/Issue : 2026;03(04)

Page No : 1000-1014

A Secure Mobile Voting System in a College is a digital platform that enables students and staff to cast their votes electronically using mobile devices such as smartphones or tablets. The system is designed to replace or complement traditional paper-based voting methods by providing a faster, more efficient, and user-friendly approach to conducting elections within the campus environment. The core objective of this system is to ensure security, transparency, and accuracy throughout the voting process. It incorporates advanced security mechanisms such as user authentication (via login credentials.

Keywords :

Author : 1Mr.R.KrishnaNayak,2M.Varshitha,3K.Pravalika,4I.Harshini,5G.Spandana

Title : Secure Mobile Voting In College

Volume/Issue : 2026;03(04)

Page No : 1015-1019

Educational institutions receive a large number of admission-related queries from students and parents every academic year, including information about courses, eligibility criteria, fee structure, admission procedures, and important deadlines. Traditional methods such as phone calls, emails, and manual help desks are often time-consuming, inefficient, and limited by staff availability. During peak admission periods, these approaches can lead to delayed responses, inconsistent information, and increased workload on administrative staff, ultimately affecting the overall student experience and decision-making process. To overcome these challenges, an AI-Based Admission Query Support and Assistance System is proposed, which uses Natural Language Processing (NLP) and Machine Learning techniques to automatically understand and respond to user queries in real time. The chatbot operates 24/7, handles multiple users simultaneously, and provides instant, accurate, and consistent information through an interactive platform.

Keywords :

Author : Mrs. A. Eenaja1 ,P. Keerthi Reddy², R. Kalyani³, M.Sravanthi⁴, B.Ramya⁵

Title : INTERACTIVE CHATBOT CREATION FOR COLLECTING THE ADMISSION QUERIES

Volume/Issue : 2026;03(04)

Page No : 1020-1025

In the modern digital era, communication and documentation have largely shifted from traditional handwritten methods to digital text due to its speed, convenience, and ease of storage. Despite these advantages, digital text often lacks the personal touch, uniqueness, and emotional connection that handwritten content naturally provides. To bridge this gap, the Smart Handwriting System is developed as an innovative web-based solution that enables two-way conversion between handwritten and digital text. The system allows users to convert typed text into realistic handwritten content, while also enabling users to upload handwritten images or PDF files and convert them into editable digital text. Developed using HTML, CSS, JavaScript, and Python (Flask), the system provides high-quality output in image and PDF formats with an intuitive, user-friendly interface. Features such as file upload and conversion history further enhance usability, making it a practical tool for students, professionals, and creative applications.

Keywords :

Author : 1Mr. R. Krishna Nayak, 2B. Sravani, 3CH. Rakshita, 4G. Sanjana, 5B. Deepthi Nagamai

Title : Smart Hand Writing System

Volume/Issue : 2026;03(04)

Page No : 1026-1029

Concrete is always expected to be stronger and more durable than in the past while being cost and energy efficient. Moreover the major advantages that concrete possesses over the construction materials have to be conserved. The possibility of being fabricated practically anywhere, the ability to make the form imposed by the shape of a mould and a low cost of components and manufacture. These factors have driven advances in improving the performance of concrete over years and continue to do so the need for improving the performance of concrete and concern for the environmental impact arising from the continually increasing demand for concrete has lead the growing use of alternative material components. An experimental investigation will be conducted to study the properties of concrete containing Fly ash, Rice Husk ash and Egg shell powder as a partial replacement of cement in the concrete mix design. Various Strength tests compressive strength, split tensile strength, flexural strength and durability of concrete by varying proportions of RHA and ESP with constant variation 5% of FA. The experimental trials used for this study are 0%FA+0%RHA+0%ESP, 5%FA+0%RHA+0%ESP, 5%FA+5%RHA+5%ESP, 5%FA+10%RHA+10%ESP, 5%FA+15%RHA+15%ESP by weight of cement for M30 grade concrete . The obtained results will be compared with the conventional concrete, there by knowing the changes in the properties of concrete containing FA, RHA, and ESP as partial replacement of cement.

Keywords :

Author : DASARI TEJASWINI, C UMAMAHESWAR

Title : STUDY ON STRENGTH AND DURABILITY OF CONCRETE BY USING FLY ASH RISE HUSK ASH AND EGG SHELL POWDER

Volume/Issue : 2026;03(04)

Page No : 1030-1036