Journal of Data Engineering and Knowledge Discovery

Analytical Study of Social Media Sentiment and Engagement Pattern using Machine Learning Approach

2026-07-14T10:16:40+00:00

Nowadays, millions of posts are published on well-known social media platforms by users. Every post that is input by the user has different reactions in terms of likes, shares, and comments from people. Some posts have created a greater impact on people, and some have not. But some sectors like business and marketing, politics, news, etc. are eager to take advantage of social media posts. Priority-wise, they are thinking about recording reactions by the people on multiple posts created by the user. This study helps to gain insight into people’s reactions to specific posts in terms of likes, shares and comments to know the actual sentiments of people against this post and their analysis in terms of engagement, i.e., posts of positive, negative and neutral nature for the people. This project analyzes the text of social media posts along with user engagement metrics, including likes, shares, and comments. Using machine learning and sentiment analysis techniques, the system classifies each post as positive, negative, or neutral and evaluates the corresponding user engagement.

An Intelligent Computer Vision and CNN Framework for Automated Plant Disease Detection and Precision Agriculture

2026-06-18T07:08:02+00:00

Plant diseases represent one of the most critical threats to global agricultural productivity, where maize crops are particularly susceptible to multiple fungal diseases that significantly reduce yield quality and quantity. Traditional disease identification methods depend on manual visual inspection by agricultural experts, which is time-consuming, inconsistent, and impractical for large-scale precision farming applications. Within the paradigm of intelligent agricultural systems, this study proposes an automated deep learning and computer vision framework for maize leaf disease identification and classification. The proposed system introduces a novel Multi-Resolution Weighted CNN (MRW-CNN) architecture that leverages multi-scale feature extraction to achieve superior classification performance across five disease categories, including Common Rust, Grey Leaf Spot, Healthy, Northern Leaf Blight, and Southern Rust. A comprehensive dataset of 2,500 annotated maize leaf images with disease stage annotations (early, advancing, and severe) is utilized for training and evaluation. Comparative analysis of nine fine-tuned CNN architectures demonstrates that MRW-CNN achieves the highest testing accuracy of 97.04% with validation accuracy of 96.56%, outperforming established architectures including Xception (95.80%), MobileNet (94.64%), and Inception V3 (94.48%). The experimental results confirm the effectiveness of the proposed approach in delivering accurate, scalable, and computationally efficient automated disease detection, contributing toward intelligent precision agriculture and early crop disease management systems.

Data-Driven Crop Recommendation Using Machine Learning: A Comparative Study of Classification Algorithms

2026-07-27T11:09:57+00:00

Choosing the right crop for a given field is one of the earliest and most consequential decisions a farmer makes, yet across much of the developing world it is still guided by habit and word-of-mouth rather than by measurable soil and weather evidence. This paper develops and benchmarks a machine-learning-based crop recommendation model that maps soil nutrient concentration (nitrogen, phosphorus and potassium), soil pH, ambient temperature, humidity, and rainfall onto the most suitable crop for cultivation. Six supervised classifiers Random Forest, Decision Tree, Support Vector Machine, K-Nearest Neighbours, Logistic Regression, and Naïve Bayes were trained on a benchmark agronomic dataset and compared using accuracy, precision, recall, F1-score, and learning-curve diagnostics. Random Forest achieved the best overall balance, with 93.6% accuracy and an F1-score of 0.936, closely trailed by Decision Tree and Naïve Bayes, whereas the Support Vector Machine underfit the data and reached only 62.4% accuracy. These findings indicate that tree-based ensemble learners generalise more reliably on structured agronomic data than distance-based or margin-based classifiers, and they support the choice of Random Forest as the recommendation engine of a lightweight, cloud-deployable crop-advisory tool.

PulseAI: A Real-Time Public Sentiment Analysis Framework Using Transformer-based Language Models on Customized Data

2026-06-29T06:46:33+00:00

In today’s digital world, social media platforms generate massive amounts of real-time textual data that reflect public opinions, emotions, and trends. Analyzing this data efficiently has become essential for businesses, researchers, and decision-makers to understand customer behavior and public sentiment. This project presents PulseAI, a real-time public sentiment analysis framework that uses Transformer-based Natural Language Processing techniques for intelligent opinion mining across multiple online platforms. The system integrates data from sources such as Twitter/X, Reddit, YouTube, Instagram, and news platforms using asynchronous API-based data collection mechanisms. The proposed framework employs a fine-tuned RoBERTa/BERT model for sentiment classification and contextual understanding of textual data. Unlike traditional machine learning approaches, the transformer-based architecture effectively captures semantic relationships, contextual dependencies, and sarcasm in user-generated content. The system performs sentiment classification into Positive, Negative, and Neutral categories while also providing emotion detection and aspect-based sentiment analysis. A FastAPI backend processes the incoming data efficiently, while a React-based dashboard visualizes sentiment trends, platform-wise analysis, heatmaps, word clouds, and real-time analytics. Experimental evaluation demonstrates that the proposed PulseAI framework achieves high performance with an accuracy of approximately 94%, outperforming conventional models such as Naïve Bayes and LSTM. The framework provides scalable, accurate, and real-time sentiment intelligence that can support applications in business analytics, brand monitoring, market research, and public opinion analysis. Future enhancements may include multilingual sentiment analysis and multimodal AI techniques combining text and image understanding.

Development of an Adaptive E-Learning Model for Personalized Learning Path Recommendation Using Ant Colony Optimization and Hybrid Filtering

2026-06-18T06:50:30+00:00

The majority of traditional e-learning models often fail to cater to individual learners’ needs due to limited parameter flexibility across diverse populations and the scope of adaptive courses. To address these challenges, this research developed an adaptive e-learning model using Ant Colony Optimization (ACO), collaborative filtering, and content-based filtering. The ACO was used for adapting the learning contents and activities, while also dynamically adjusting the learning path. K-Nearest Neighbor (KNN) was used for e-learning styles and teaching strategies to generate courses for learners. The Felder-Silverman Learning Style Model is employed to identify and accommodate different learning styles, ensuring a tailored educational experience. The implementation of the model is done using Python and PHP frameworks. An online survey was conducted with 500 undergraduate students from various academic disciplines to evaluate the developed model. The KNN algorithm achieved the best performance with an R² of 0.986 and an MAE of 0.012, indicating both high accuracy and minimal prediction error. A direct comparison with the benchmark study reveals that our models outperform their counterparts, with the KNN model achieving (0.012, 0.986) against theirs (0.013, 0.875). It demonstrates significant improvements in learner satisfaction and academic performance. This research contributes to e-learning by offering a flexible, adaptive solution that enhances educational outcomes through personalized learning experiences.