Academic Performance Prediction Based on Multisource, Multifeature Behavioral Data

In the current era of digital education, analysing student feedback effectively is vital for enhancing the quality of the teaching and learning process. Predicting academic performance is critical in educational data mining, as it can help identify and address students early and support personalised learning interventions. It is a study that investigates opinion mining through the application of supervised learning techniques to detect and classify sentiment in student feedback. The primary role is to understand the emotional tone of student responses and derive meaningful insights to support the same in academic settings. And also focuses on academic performance prediction using multi-source, multi-feature behavioural data collected from students across various digital platforms and learning environments. Unlike traditional approaches that rely solely on demographic or historical grade data, this research integrates Aviator's range of behavioural indicators, including attendance records, online learning activity logs, assignment submission patterns, participation in discussions, library usage and even social interactions with academic systems. The dataset used in the study comprises student feedback collected from the module evaluation service conducted at Dr. Paul Raj Engineering College, Yetapaka. The feedback includes both structured and unstructured a reflecting students’ views on teaching methodologies, learning outcomes, assessment procedures and overall course delivery. To process this data, a combination of Artificial Intelligence(AI) and Natural Learning Process(NLP) techniques has been employed. The study emphasises the use of Python, an open source programming language, along with relevant libraries such as Scikit-learn, NLTK, and Pandas, to implement the sentiment analysis model. Various supervisor learning algorithms, including Naive Bayes, Support Vector Machines(SVM) and decision trees, have been applied to classify the sentiments as positive, negative and neutral. These models are evaluated based on standard performance metrics such as accuracy, precision, recall and F1-score. The experiment results show that combining multiple source behavioural features significantly improves prediction accuracy compared to using single source data alone. The proposed model employs advanced learning techniques to process and analyse the heterogeneous data collection from multiple sources of learning management system institutional databases and campus infrastructure. This research provides valuable insights into how different aspects of student behaviour influence academic success. It also highlights the potential of data-driven decision-making in academic settings and enables educators to design more effective and personalised intervention strategies. By leveraging multiple-source multiple-feature institutions in North only academic outcomes can but also enhance overall student support services.