Feature Extraction and Prediction Based on Random Forest and BP Neural Network

This paper presents a comprehensive study on feature extraction and prediction utilizing Random Forest (RF) and Backpropagation Neural Network (BPNN). Initially, Spearman's rank correlation analysis is employed to assess the correlation between 20 indicators and the target variable. Subsequently, the Random Forest algorithm is leveraged to evaluate the importance of these 20 features in predicting the occurrence probability. Through this process, a subset of highly correlated features is identified. Building upon the RF model, the five most significant features are selected, and the refined dataset is then utilized to train a BPNN model. Evaluation metrics, including Mean Squared Error (MSE) and the Coefficient of Determination (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex>), are calculated for both training and testing sets. The results indicate that the BPNN model exhibits no signs of overfitting, explains approximately 62 % of the variance in the target variable, and demonstrates robust predictive performance.