Enhanced Firefly Algorithm with Lévy Flight Strategy for Feature Optimization in LSSVM-Based Online Learner Behavior Recognition

Abstract

With the rapid development of online education, the precise identification of learners' behavioral patterns to enhance the effectiveness of personalized teaching has become an important research topic in the field of smart education. To improve the accuracy of online learning behavior recognition, this study proposes an integrated model based on an optimized least squares support vector machine using an improved firefly algorithm. This method improves the skip search and local convergence capabilities of FA in high-dimensional feature spaces. It introduces a Lévy flight mechanism and a dynamic step-size adjustment strategy. It also combines a bagging ensemble strategy to construct a multi-subclassifier fusion structure. These improvements effectively enhance the model's generalization performance and classification stability. The experimental part is based on the UCI public student learning behavior dataset and real Moodle platform behavior logs for evaluation. The model performance was systematically tested using 50% cross validation and comparative experiments. The results showed that the proposed model achieved accuracies of 97.84% and 97.52% on the training and testing sets, respectively. The minimum classification error rate was 2.42%. In terms of error metrics, the proposed model outperformed others in classification error rate, root mean square error, and cross-entropy loss. Specifically, the classification error rate for problem-solving tasks was 3.15%, the root mean square error was 0.10, and the cross-entropy loss was 0.23. Meanwhile, the model has good resource utilization control, with an average memory usage of less than 450MB and a CPU usage rate of less than 65%. The proposed model demonstrates high accuracy and scalability in recognizing multi-class behaviors. It is suitable for automating the modeling process and deploying intelligent teaching support systems for large-scale learning behavior data on online educational platforms.