Background

Addressing the varying learning styles of students in STEM education remains a significant challenge, often resulting in reduced engagement and inconsistent comprehension. Personalized teaching strategies can effectively bridge these gaps. However, most traditional models lack real-time adaptability, limiting their ability to dynamically adjust to evolving student needs.

Objective

This study focuses on developing a multi-agent framework capable of analyzing and categorizing individual learning preferences—Visual, Auditory, and Kinesthetic—to deliver customized educational experiences that enhance student engagement and outcomes. Additionally, the framework integrates real-time adaptive learning by incorporating reinforcement learning models and continuous feedback mechanisms to dynamically refine teaching strategies.

Methodology

The proposed framework consists of a Teacher Agent, Concept Mapping Agent, Content Analysis Agent, and Sentiment Analysis Agent. Using advanced machine learning algorithms, including Multinomial Naive Bayes, Support Vector Machine (SVM), Decision Tree (DT), and Random Forest (RF), the system classifies learning styles. A reinforcement learning model is introduced to continuously adjust content delivery based on student interactions and engagement levels. The Sentiment Analysis Agent provides real-time feedback, enabling the system to adapt its instructional approach dynamically. Evaluation metrics such as Precision, Recall, Accuracy, and F1-score ensure robust performance analysis.

Results

The system demonstrated an outstanding classification accuracy of 98%, with Multinomial Naive Bayes leading in performance. The integration of real-time sentiment analysis significantly improved engagement tracking, allowing the framework to refine content delivery based on student responses. The reinforcement learning model successfully adjusted teaching strategies dynamically, leading to improved learning outcomes over time.