Multi-objective Prediction Model based on GA-BP Neural Network and Logistic Regression

Abstract

This study develops a multi-objective prediction model integrating two complementary approaches to address complex prediction tasks in hierarchical data structures. The first is a hybrid Genetic Algorithm-Back Propagation Neural Network (GA-BP), which utilizes advanced feature selection techniques, including Lasso regression and XGBoost, to identify key predictors while addressing nonlinear dependencies and convergence issues. The GA-BP model achieves enhanced robustness, effectively modeling complex relationships through global optimization of initial weights and biases. The second approach is a logistic regression submodel, designed to estimate probabilities of binary classification events with a focus on refined feature adjustments and regularization techniques. This framework highlights the significance of feature engineering, integrating both objective and subjective feature weightings to improve accuracy across multivariable datasets. By combining machine learning methodologies with statistical rigor, this integrated model enhances prediction performance and provides actionable insights for diverse use cases in complex systems.