Construction of Student Performance Prediction Model Based on Data Mining and Optimized RBF neural Network

Abstract

K-means cluster analysis algorithm and RBF neural network (RBFNN) are important tools for predicting students' final grades. Aiming at the defect that the K-means needs to manually select the K value, and the efficiency and accuracy are low, the research uses the idea of DBSCAN to optimize it to extract the characteristics of students' learning behavior. Because the low efficiency of manual setting of RBF neural network parameters, and the setting results are often not optimal parameters, the research uses the improved K-means and equalized discriminant function to determine the network center point and the parameter of the RBFNN. Optimization of RBFNN. The learning behavior features extracted by the improved K-means are input into the improved RBFNN, and the students' final grades are predicted according to the output results. Based on the above content, the research builds a performance prediction model based on the improved RBFNN. The model 1 only needs 58 times to achieve the target accuracy, the fitting accuracy reaches 0.957, the accuracy rate reaches 99.2%, and the average prediction error rate is 0.03. It shows that the grade prediction model constructed by the research can predict students' final grades with high precision and high efficiency, which will help teachers obtain more intuitive feedback, thereby improving teaching methods and improving students' academic performance.