Research on Prediction Methods Based on Multiple Linear Regression - Artificial Neural Network Model

Abstract

The study employs a predictive framework based on a multiple linear regression-artificial neural network (MLR-ANN) model to achieve high-precision predictions for specific targets. First, features are extracted from both overall and individual dimensions, followed by data processing. An artificial neural network model is constructed to uncover non-linear features within the data. Second, a multiple linear regression model is used to fit linear relationships, while incorporating probability distribution characteristics to quantify uncertainty in the prediction process. A result aggregation strategy is further employed to integrate the two models, forming a comprehensive prediction system. Additionally, the study uses a logistic regression model to analyse the inter-variable association mechanisms and evaluates the importance of specific factors through the EWM-TOPSIS method. This framework enhances prediction accuracy and reliability through multi-dimensional feature analysis and model optimisation, providing an analytical framework and methodological reference for similar problems.