VOLUME 15, ISSUE 10, OCTOBER 2026

Abstract: Ransomware remains one of the major cybersecurity threats because modern ransomware variants continuously evolve and may differ considerably from previously identified samples. Traditional signature-based detection approaches are effective against known threats but may have limited capability against previously unseen variants. Machine learning provides an alternative approach by learning discriminative patterns from malicious and benign executable samples. However, high classification performance under a conventional random train-test split does not necessarily demonstrate generalization to ransomware families that were not represented during training.

This research proposes an AI-Based Ransomware Detection Framework Using Machine Learning based on static characteristics of Windows Portable Executable (PE) files. The framework extracts a compact set of eight static features consisting of API count, DLL count, unique API count, unique DLL count, file size, packed status, entropy, and PE timestamp year. Five machine learning algorithms—Logistic Regression, Decision Tree, Random Forest, XGBoost, and LightGBM—are trained and compared using accuracy, precision, recall, F1score, ROC-AUC, false positive rate, and false negative rate.

The experimental framework additionally incorporates validation-based probability threshold analysis, ransomware- family-disjoint evaluation, family-level generalization analysis, Isolation Forest-based novelty detection, and SHAP- based explainability. Random Forest achieved the best overall standard-test performance with 95.42% accuracy, 96.22% precision, 94.28% recall, 95.24% F1-score, and 98.98% ROC-AUC. When evaluated on ransomware families excluded from training, the Random Forest model achieved 90.20% accuracy, 98.88% precision, 87.02% recall, 92.57% F1-score, and 97.72% ROC-AUC. The family-disjoint experiment demonstrates that detection becomes more challenging when the model encounters ransomware families not represented during training. Isolation Forest additionally identified approximately 8.04% of unseen ransomware samples as anomalous relative to the known ransomware distribution.

The results demonstrate that a compact static PE representation combined with ensemble machine learning can provide strong ransomware classification performance while enabling additional analysis of generalization, novelty, and interpretability.

Keywords: Ransomware Detection, Machine Learning, Static Analysis, Portable Executable, Random Forest, Explainable AI, SHAP, Unseen Ransomware Families, Anomaly Detection, Malware Detection.

Abstract: Government provide many schemes for farmer.the scheme is on official portal,but most of the scheme are in English .In Maharashtra many Farmers know only Marathi language,they did not aware about English language,due to this they are not able to take the benefits of the governments schemes.
My Research objectives is ,all farmers can get benefited from governments schemes,they can get advantage of the government policies also. I am used Experimental Methodology and I am used A Transformer-based Neural Machine Translation .the performance of model was evaluated using BLEU and CHRF metrics.the system helps to farmers for easily understand government scheme information,specially for Marathi speaking farmers.In future I am focusing on expanding the Data set of farmer related schemes,supporting additional Indian languages and voice based interaction.

Keyword: Natural Language Processing(NLP),English-Marathi Translation ,Transformer,Neural Machine Translation,Government Schemes,Agricultural Information,Marathi Language

Abstract: This study analyzed institutional enrollment trends across six academic years, from Academic Year (AY) 2018–2019 to AY 2023–2024, with emphasis on changes by college and sex. Specifically, it compared the combined enrollment counts during the first three academic years (2018–2019 to 2020–2021) with those during the last three academic years (2021–2022 to 2023–2024). A descriptive-comparative approach was used to analyze institutional enrollment records. Frequency counts, percentages, numerical differences, and percentage changes were employed to identify patterns of enrollment growth and decline. Findings revealed that annual enrollment initially decreased by 5.34% from AY 2018–2019 to AY 2019–2020 but subsequently recovered, with the largest annual increase of 26.97% occurring in AY 2021–2022. Enrollment reached a six-year high of 15,043 in AY 2023–2024. Comparing the two three-year periods, overall enrollment increased from 29,901 to 42,231, representing an increase of 12,330 or 41.24%. Seven of the eight colleges recorded enrollment growth, while the College of Education (COED) decreased by 3.73%. The College of Nursing (CON) demonstrated the largest percentage increase at 157.23%, followed by the College of Arts and Sciences (CAS) at 44.15% and the College of Allied Medical Sciences (CAMS) at 41.05%. Female enrollment remained higher than male enrollment, accounting for 61.30% of enrollment during the last three-year period. Female enrollment increased by 48.81%, compared with 30.70% among males. However, sex composition differed across colleges, with males remaining the majority in CCJE and CECT. The findings demonstrate substantial institutional enrollment expansion, although the magnitude and sex distribution of growth varied across academic units. These enrollment analytics provide an empirical and data-driven basis for institutional planning and for advancing the digitalization and automation of registrar operations, particularly in enrollment monitoring, automated reporting, data visualization, and decision support. Integrating historical enrollment data into a digital registrar information environment may facilitate more efficient monitoring of enrollment patterns, resource allocation, staffing, student services, and program development. The figures represent enrollment counts across the first semester, second semester, and summer terms and should not be interpreted as counts of unique individual students.

Keywords: academic programs, enrollment trends, higher education, institutional planning, sex distribution

Abstract: Stock prices are influenced not only by historical price movement but also by public sentiment expressed through financial news. Retail investors usually consult charting platforms and news platforms separately, with no automated bridge between them. This project presents an implemented, Flask-based web application that combines quantitative technical analysis of stock prices with machine-learning based sentiment analysis of financial news to produce a transparent BUY, HOLD or SELL recommendation.

Historical price data is sourced from Yahoo Finance and used to calculate the 20-day moving average, the 14-period Relative Strength Index (RSI) and the Moving Average Convergence Divergence (MACD). Financial news is gathered from NewsAPI, cleaned with NLTK, and converted into TF-IDF and hand-crafted sentiment features. Three classifiers, namely Random Forest, Linear Support Vector Machine and Multinomial Naive Bayes, are trained, and the best is selected by test-set accuracy. Article-level predictions are aggregated into a daily Sentiment Index, which is combined with a price-trend score using the weighted formula 0.6 x Sentiment + 0.4 x Trend. An RSI-based safety filter prevents extreme-market signals from being followed blindly. All data and results are stored in SQLite and displayed on an interactive Chart.js dashboard.

The system was run on 411 stored news articles covering seven US and Indian stocks. On an 80/20 split, the Linear SVM achieved the highest test accuracy (100.00%), followed by Random Forest (98.80%) and Naive Bayes (79.52%), and was selected automatically. Because the sentiment labels are generated by a financial keyword lexicon rather than by human annotators, these metrics measure agreement with that labelling heuristic and not validated real-world sentiment accuracy. This limitation is noted explicitly throughout the report.

Keywords: stock market, sentiment analysis, technical indicators, RSI, MACD, TF-IDF, Random Forest, SVM, Naive Bayes, Flask.

Abstract: A major aspect of determining student attendance is having accurate records; unfortunately, traditional attendance systems are inefficient, not accurate, and can be manipulated for ”proxy” attendance. A way around these obstacles is to create an intelligent, automated attendance method using deep learning facial recognition technology based on a pretrained model. In addition to traditional still-image approaches, our system begins with recording a short video of each participating student, leading to frame-by-frame extraction for constructing a robust and diverse dataset containing facial image samples of each student. This new pipeline from video to image will improve the representation of students’ faces by using the three ways students’ faces are captured (pose, expression, and lighting) to provide a more realistic photo of each student. In this approach, we will utilise a pretrained deep learning face recognition algorithm to extract the distinctive features of student faces from our generated dataset. A single classroom image containing multiple students will be used for attendance. The system will identify faces in the group image and extract their unique features by using three face detection methods, namely Haar Cascade, MTCNN, and InsightFace, which are evaluated and compared. For recognition, FaceNet and ArcFace embedding models are investigated using cosine similarity matching to perform both extraction and recognition, which helps us generate face embeddings for each person, compare those features to the database using cosine similarity, and mark attendance for recognized students with no manual intervention. This approach should successfully connect individual-level training data with group-level testing conditions, allowing for accurate facial recognition in the classroom.

Keywords: Automated Attendance System, Facial Recognition, Pretrained Deep Learning Model, Video Frame Extraction.

Abstract: Tamil-English code-mixing, popularly known as “Tanglish,” is a pervasive mode of informal digital communication among Tamil speakers, frequently rendered in Romanized script across social media and messaging platforms. Existing natural language processing (NLP) systems, largely trained on monolingual or formally structured corpora, are not optimized to interpret the semantic, sentiment, and intent-level nuances of such hybrid, non-standardized text. This paper proposes a comparative research framework to evaluate the capability of different computational approaches—traditional NLP techniques, BERT-based transformer models, and large language models (LLMs)—in understanding Tamil-English code-mixed text, alongside a proposed preprocessing pipeline designed to normalize Romanized Tamil variants prior to model inference. We outline a methodology for constructing an annotated code-mixed dataset, detail the architecture of each comparative model, and define evaluation metrics spanning accuracy, F1-score, and semantic similarity. The study aims to establish whether targeted preprocessing can meaningfully improve semantic fidelity for low-resource, code-mixed language understanding, contributing toward AI systems that better preserve and process minority and regional language expression in digital spaces.

Keywords: code-mixing, Tanglish, Tamil-English NLP, Romanized Tamil, semantic preservation, low-resource language processing, transformer models

Abstract: Effective waste classification is a critical component of modern smart waste management systems, enabling automated sorting and recycling at scale. This paper presents a comprehensive comparative study of six convolutional neural network (CNN) architectures (VGG16, ResNet50, MobileNetV3, EfficientNet-B0, AlexNet, and a custom- designed CNN) for the task of automated waste classification into six categories: cardboard, glass, metal, paper, plastic, and trash. Each model is evaluated on a standardized dataset of 2,527 images under identical experimental conditions, including data augmentation, learning rate scheduling, and transfer learning where applicable. Performance is assessed using accuracy, precision, recall, F1-score, receiver operating characteristic (ROC) curves, and precision-recall (PR) curves. Experimental results demonstrate that the custom CNN achieves the highest classification accuracy of 95.05% with a macro-average ROC-AUC of 0.9974, outperforming established architectures. ResNet50 and MobileNetV3 follow closely at 94.07% and 93.87%, respectively, offering favorable trade-offs between accuracy and computational complexity. The analysis further includes per-class performance evaluation, confusion matrix analysis, and computational efficiency comparison, providing actionable insights for deploying waste classification systems in resource-constrained environments.

Keywords: waste classification, convolutional neural networks, transfer learning, deep learning, image classification, recycling, smart waste management

Abstract: Mental stress is becoming a common issue among students and individuals due to factors such as workload, lack of sleep, screen time, deadline pressure, financial concerns, personal responsibilities, and limited relaxation time. Identifying stress levels can be difficult because different factors can affect each individual differently. This research presents a machine learning framework for the detection of mental stress using questionnaire-based data. Data was collected from 51 respondents through a Google Form and included factors such as sleep hours, screen time, workload, physical activity, socialization, sleep quality, deadline pressure, financial concerns, personal or family responsibilities, and social support. The collected data was processed and used to calculate a stress score, based on which respondents were classified into Low, Moderate, and High stress levels. The dataset contained 17 columns and the stress-level distribution included 10 Low, 37 Moderate, and 4 High stress responses. A machine learning classification model was developed using Python to identify stress-level patterns. The model achieved an accuracy of approximately 81.82%. The study also used data visualization to understand the distribution of stress and the relationship between different factors and stress levels. The proposed framework shows how data analysis and machine learning can be used as a preliminary approach for detecting and understanding mental stress patterns.

Keywords: Mental Stress, Machine Learning, Stress Detection, Stress Level, Data Analysis, Python, Classification

Abstract: Leukemia is a life-threatening hematological malignancy where early and accurate detection significantly improves patient survival. Conventional diagnosis relies on manual microscopic examination of peripheral blood smears, a process that is time-consuming, observer-dependent, and prone to error. This paper presents HemaSight AI, a four- stage artificial intelligence pipeline for automated leukemia screening and post-transplant monitoring. Stage 1 applies XGBoost to classify complete blood count (CBC) parameters, including 17+ extended markers such as SGOT, SGPT, creatinine, potassium, chimerism, and GVHD status. Stage 2 employs a conditional generative adversarial network (cGAN) to synthesize microscopic blood smear images alongside a ResNet18-based similarity engine that computes a weighted four-metric match (semantic, color, structural, pixel). Stage 3 uses linear regression on longitudinal blood test data to detect early signs of relapse before clinical symptoms appear. Stage 4 integrates Google Gemini via a multilingual chatbot supporting seven languages. The platform is implemented using Next.js, FastAPI, Supabase, and PyTorch, deployed entirely on free-tier infrastructure. Evaluation across 68 test cases achieved a 98.5% pass rate with sub-3-second API response times. Results demonstrate that a low-cost, scalable, and clinically accessible AI tool can be built without proprietary infrastructure.

Keywords: Leukemia detection, XGBoost, conditional GAN, ResNet18, time-series analysis, large language model, medical AI, multilingual chatbot.

Abstract: Smart Irrigation System for Precision Farming is an IoT-based and sensor-driven approach designed to improve water management, crop productivity, and agricultural sustainability. Conventional irrigation practices often depend on fixed schedules or manual decisions, which can result in over-irrigation, under-irrigation, water wastage, and increased labour requirements. The proposed system integrates soil-moisture, temperature, humidity, and rainfall sensors with a microcontroller such as Arduino or ESP32 to continuously monitor field conditions in real time. Based on the collected data, the system automatically determines the irrigation requirement and controls water pumps or solenoid valves accordingly. Weather information can also be incorporated to avoid unnecessary irrigation during expected rainfall and to adapt watering according to environmental conditions. IoT connectivity enables farmers to remotely monitor soil and weather parameters through a web or mobile dashboard. The system provides a low-cost, scalable, and farmer-friendly solution that supports precision agriculture by delivering the right amount of water at the right time. Such intelligent irrigation can conserve water and energy, reduce manual effort, improve crop health, and contribute to sustainable agricultural development.

Abstract: Crop diseases are a major challenge in agriculture, as late or incorrect identification can cause crop damage and reduce productivity. Existing crop disease detection systems mainly focus on identifying diseases from plant leaf images, but provide limited support for disease management. This creates a gap between disease identification and practical management. The proposed Multi-Crop Disease Detection and Management Recommendation System Using Deep Learning addresses this gap by combining disease detection with management recommendations. The system uses deep learning and transfer learning models to classify crop diseases from leaf images. It provides chemical and organic management approaches based on the identified crop and disease. Thus, the system connects disease identification with suitable management practices, making it more useful for practical crop disease management.

Keywords: Crop Disease Detection, Deep Learning, Management Recommendation System, Multi-Crop Classification, Transfer Learning.

Abstract: Digital documents are widely used in education, banking, healthcare, government, employment, and business. Ensuring the authenticity and integrity of these documents has become increasingly important due to the growing number of forged, altered, and fraudulent documents. Traditional verification methods rely heavily on manual inspection and comparison with original records, which can be time-consuming, costly, and prone to human errors. Recent developments in Artificial Intelligence (AI), Machine Learning (ML), Deep Learning (DL), and Optical Character Recognition (OCR) have enabled automated document verification with improved accuracy and efficiency. Machine learning algorithms such as Random Forest, Support Vector Machine, Decision Trees, Logistic Regression, and K-Nearest Neighbors can classify genuine and fraudulent documents using extracted features. Deep learning models such as Convolutional Neural Networks (CNNs) can analyze document images and detect manipulation and forgery. This review examines recent advancements in digital document verification, including OCR, image processing, digital signatures, blockchain, biometric verification, cloud computing, and multimodal document analysis. The study also discusses commonly used datasets, verification methods, evaluation metrics, challenges, research gaps, and future research directions. The findings indicate that integrating AI-based verification, secure digital signatures, blockchain, and multimodal analysis can support the development of reliable, secure, scalable, and efficient digital document verification systems

Keywords: Digital Document Verification, Document Authentication, Machine Learning, Deep Learning, Optical Character Recognition, Artificial Intelligence, Blockchain, Digital Signature, Forgery Detection.

Abstract: Customer complaints provide valuable information about problems in products, services, delivery, support, and customer experience. However, organizations often receive a large number of complaints through websites, email, social media, call centers, and mobile applications, making manual processing slow and difficult to manage. A customer complaint tracking system can improve this process by automatically collecting, classifying, prioritizing, routing, monitoring, and resolving complaints. This paper presents a review of recent approaches based on Natural Language Processing (NLP), Machine Learning (ML), Deep Learning (DL), sentiment analysis, topic modelling, and transformer-based models for intelligent complaint management. NLP techniques can convert unstructured complaint text into useful information, while machine learning algorithms can classify complaints according to category, urgency, department, or severity. Sentiment analysis can identify customer emotions and support priority assignment, whereas automated routing can forward complaints to the appropriate department. Recent research has also explored BERT and AI-enabled complaint-handling systems. The review identifies important research gaps related to multilingual complaints, limited labelled datasets, explainability, privacy, real-time tracking, integration with existing business systems, and human-AI collaboration. Future systems should combine NLP, explainable AI, large language models, sentiment analysis, automated escalation, and real-time dashboards to provide faster and more transparent complaint resolution.

Keywords: Customer Complaint Tracking, Complaint Management, Natural Language Processing, Machine Learning, Sentiment Analysis, Text Classification, BERT, Topic Modelling, Automated Routing, Customer Service.

Abstract: Agriculture plays an important role in food security and the economic development of many countries. Selecting the most suitable crop for a particular field is a major challenge for farmers because crop growth depends on soil properties, nutrients, temperature, humidity, rainfall, pH, and climatic conditions. Traditional crop selection mainly depends on farmers experience and local knowledge, which may not always provide accurate recommendations under changing environmental conditions. Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) provide effective solutions for data-driven agricultural decision making. This paper presents a review of crop recommendation systems with particular emphasis on Convolutional Neural Networks (CNNs). CNNs can automatically learn useful patterns from soil images, remote-sensing images, and other agricultural data, while CNN- based hybrid models can also be combined with environmental and soil parameters. Recent studies have explored Random Forest, XGBoost, Support Vector Machine, LSTM, CNN, attention-based networks, Transformer models, and hybrid deep-learning architectures for crop recommendation. This review examines research published from 2020 to 2026, summarizes their datasets, preprocessing methods, algorithms, and major findings, and identifies research gaps related to dataset availability, regional generalization, explainability, multimodal data integration, and real-time deployment. A CNN-based crop recommendation system can potentially provide farmers with localized and intelligent recommendations while reducing inappropriate crop selection, improving productivity, and supporting sustainable farming.

Keywords: Crop Recommendation, Convolutional Neural Network, CNN, Deep Learning, Precision Agriculture, Smart Farming, Soil Classification, Machine Learning, Farmers, Agriculture.

Abstract: This research presents a non-linear tumor-immune-chemotherapy model that distinguishes drug sensitive from drug resistant tumors with respect to chemotherapy dynamics and immune response. The model includes Gompertzian tumor growth, immune mediated killing, differential drug sensitivity, treatment induced drug resistance, immune suppression, and drug clearance. Biologically feasible equilibrium states were established by analytical investigation along with biological conditions of threshold for invasion by sensitive or resistant tumor cell types. Chemotherapy is demonstrated to have a double-edged sword effect in this model; increasing treatment intensity will be effective at suppressing sensitive tumor cells, however it also promotes the relative proportion of resistant tumor cells. Numerically simulated results demonstrate that local stability remains around positive equilibria and resistant tumor cells can remain present within the model despite an increase in chemotherapy. Further sensitivity analysis was conducted to identify factors that influence the models behavior including tumor growth rate, immune killing rates, chemotherapy efficacy, selective pressure on resistance development, degree of immune suppression and drug clearance. Ultimately the data supports that reduction in sensitive tumor population through chemotherapy does not eliminate all cancer (resistant cells), nor prevent recurrence due to continued presence of resistant cells.

Keywords: Tumor–immune dynamics, chemotherapy, drug resistance, mathematical modeling, equilibrium analysis, threshold conditions, Gompertz growth, resistance invasion