Research & Review: Machine Learning and Cloud Computing (e-ISSN: 2583-4835)

AgriSense: Harnessing Machine Learning for Intelligent Crop Recommendation and Sustainable Agriculture

Md. Momenul Haque, Md. Fatin Nibbrash Nakib, Mehedi Hasan — Mon, 23 Feb 2026 00:00:00 +0000

Agriculture plays a crucial role in economic development, especially in countries such as Bangladesh and India where a large proportion of the population depends on farming for their livelihood, and with increasing climate variability and resource constraints, intelligent data driven approaches are essential to improve crop productivity, sustainability, and efficient resource utilization; this study presents AgriSense, a machine learning based framework for intelligent crop recommendation that assists farmers in selecting suitable crops by analyzing rainfall patterns, climatic conditions, and soil fertility, addressing the limitations of traditional farming practices that often lack precision and lead to inefficient resource use and suboptimal crop choices; the dataset was constructed by integrating multiple publicly available datasets related to rainfall, climate, soil nutrients, and fertilizer usage specific to India, providing a rich and diverse foundation for model training and evaluation; four machine learning algorithms, Random Forest, Logistic Regression, Support Vector Machine, and Decision Tree, were applied for crop prediction, with K fold cross validation used to ensure model reliability and generalization, and performance evaluated using confusion matrices and ROC curves, the experimental results demonstrate that the Random Forest model achieved the best performance with 99.59% accuracy, 99.62% precision, 99.59% recall, a 99.59% F1 score, and 100% AUC, highlighting the effectiveness of AgriSense as a practical, scalable, and data driven solution for supporting sustainable agricultural decision making and transforming traditional farming into a more resource efficient and intelligent process.

A Smart Learning Analytics Model for Dropout Risk Estimation Using Classification Techniques

Pranita Pramod Patil — Fri, 27 Feb 2026 00:00:00 +0000

Student dropout remains a significant concern in modern education systems, affecting institutional performance, learner progression, and socio-economic development. The expansion of digital learning environments has generated large volumes of educational data, enabling data-driven approaches for predictive analysis. This study presents an intelligent predictive framework for early student dropout risk detection using machine learning classification techniques. The proposed system integrates academic performance indicators, engagement metrics from Learning Management Systems (LMS), and demographic attributes to construct a comprehensive predictive model. Data preprocessing techniques including missing value handling, normalization, and categorical encoding are applied to enhance data quality. Multiple classification algorithms such as Logistic Regression, Decision Tree, Random Forest, Naïve Bayes, and Support Vector Machine are implemented and evaluated using performance metrics including accuracy, precision, recall, and F1-score. Experimental findings indicate that ensemble-based models, particularly Random Forest, demonstrate improved predictive accuracy and stability compared to traditional classifiers. The developed framework enables early identification of at-risk students, supporting proactive intervention strategies. The results highlight the effectiveness of learning analytics and machine learning in transforming educational data into actionable insights for improving student retention and academic success.

Review of Twitter Sentiment Analysis using Deep Learning Approaches

Hemlata Pawar, Nitya Khare, Swati Khanve — Mon, 23 Mar 2026 00:00:00 +0000

Twitter has emerged as one of the most popular social media platforms where users frequently share their opinions, emotions, and perspectives on a wide range of topics, including politics, products, events, and social issues. Analyzing these opinions is important for understanding public sentiment and assisting decision-making processes across various domains. Sentiment analysis, also referred to as opinion mining, is a natural language processing (NLP) technique used to detect and categorize the sentiment expressed in textual data, typically classifying it as positive, negative, or neutral. Social media platforms such as Twitter produce a vast amount of user-generated content that captures public opinions, emotions, and attitudes regarding events, products, and policies. Sentiment analysis of Twitter data has become an important research area for understanding public perception in real time. This study conducts an empirical evaluation of Twitter sentiment classification using deep learning techniques. Several deep learning models, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Bidirectional LSTM (BiLSTM), are applied to preprocessed Twitter datasets. These models are trained to categorize tweets into positive, negative, and neutral sentiment classes. The experimental results indicate that deep learning methods outperform conventional machine learning approaches by better capturing contextual and semantic features within textual data. The results emphasize the capability of deep learning techniques to enhance the accuracy and robustness of sentiment classification in large-scale Twitter data analysis.

An Explainable Ensemble Machine Learning Framework for Early-Stage Alzheimer’s Disease Detection Using Clinical Data

Mon, 23 Mar 2026 00:00:00 +0000

Alzheimer's Disease (AD) is a progressive neurodegenerative disorder that is very important to cognitive functioning and quality of life, and therefore, early diagnosis is essential in effective management of the disease and its treatment. In this paper, an explainable ensemble machine learning model is proposed to detect early-stage Alzheimer 4 disease based on structured clinical information. They use a publicly accessible Alzheimer's clinical dataset, which is available on Kaggle, and it includes demographic features, cognitive assessment data (MMSE and CDR), as well as other clinical features. The suggested methodology is that the data is pre-processed98, features are selected, and several base classifiers such as Logistic Regression, Support Vector Machine (SVM), Random Forest, and Extreme Gradient Boosting (XGBoost) are trained. These models are combined together in terms of a stacking-based ensemble approach to provide greater classification strength and predictive capability. Shapley Additive Explanations (SHAP) are added to provide transparency and clinical interpretability, both at the global and patient level of explanation. Experimental comparison shows that the ensemble structure is more accurate, precise, recalls, and F1-score, as well as ROC-AUC, than individual models, which are reliable and interpretable decision-support systems to identify early-stage Alzheimer disease. The proposed solution is a cost effective (scalable) and clinically significant solution to cognitive impairment screening.

Deep Learning Powered Software Bug Prediction using CNN and LSTM Hybrid Architecture

Ramya P, Sivaranjani B — Mon, 23 Mar 2026 00:00:00 +0000

The rapid growth of modern software applications has significantly increased the complexity of software systems. As projects become larger and more sophisticated, identifying defects in source code has become a major challenge for developers and software engineers. Early detection of software bugs is essential because it helps reduce development costs, improves system reliability, and ensures better software quality. Traditional defect prediction approaches mainly rely on manually designed code metrics and rule-based analysis techniques. Although these methods can identify certain types of errors, they often fail to capture deeper structural and contextual relationships within the code. To address these limitations, this study introduces DeepGuard AI, an intelligent defect prediction framework based on a hybrid deep learning architecture that combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. The proposed system is designed to automatically learn meaningful patterns from raw source code without relying solely on manually defined features. The framework includes multiple stages such as data preprocessing, tokenization, embedding, feature extraction, sequential modeling, classification, and result visualization. By integrating these components into a single pipeline, the system can support efficient and accurate bug prediction. Experimental results indicate that the proposed model achieves better predictive performance than several traditional machine learning approaches used in software defect prediction.