Predicting Socioeconomic Inequality Using Machine Learning: A Study on Multi-Source Big Data Fusion Model

Abstract

Socioeconomic inequality hinders sustainable development, and its prediction is of great significance for formulating governance policies. Current studies have limitations such as single-source data and models that struggle to capture temporal dependencies. This study focuses on county-level administrative units, integrates four types of data—non-target government affairs data, social media data, remote sensing data, and mobile payment data—to construct a dataset with 776 valid records, using the county-level Gini coefficient as the target variable. After data preprocessing and feature selection, an Attention-LSTM model is built for prediction. Experimental results show that the Attention-LSTM model performs well, with a Root Mean Square Error (RMSE) of 0.021, a Mean Absolute Percentage Error (MAPE) of only 7.7131%, and a coefficient of determination (R²) of 0.8497 on the test set, which is significantly better than traditional machine learning models such as XGBoost, LSTM, and Random Forest (RF). The multi-source data fusion framework and Attention-LSTM model constructed in this study provide a technical means for monitoring socioeconomic inequality at the county level, helping policymakers identify key intervention areas and formulate differentiated governance policies.