Research on time series change point detection and influencing factors under machine learning: based on PM2.5 concentration data in Hefei city

To analyze the temporal variation characteristics of PM2.5 concentration and its influencing factors in Hefei, this paper utilizes mixed frequency data of air pollution and socio-economic development from 2013 to 2021, and constructs an adaptive change point detection model and fusion of Lightgbm in Machine Learning Models and LGBNN in multilayer neural networks to examine the change point characteristics and significant factors affecting PM2.5 concentration changes in Hefei. The findings indicate evident periodic oscillation patterns in PM2.5 concentration in Hefei, with a greater number of decreasing variables compared to rising variables in the sequence, and the change points exhibit a distinct “phased” characteristic. Regarding the influencing factors, the feature selection analysis conducted on multiple machine learning models like LGBNN reveals that policy factors exhibit the highest prominence, followed by social economy development, air pollution, and meteorological factors.