A new weighted rough set and improved BP neural network method for predicting forest fires

Abstract

To solve the quality problems of redundant risk elements, data imbalance, and noisy samples, which are commonly found in forest fire datasets, and to further improve the accuracy of forest fire risk prediction. In this paper, a forest fire prediction method is proposed, which combines a probability-weighted rough set attribute reduction (PWRS-AR) strategy with a particle swarm optimization improved BP neural network (PSO-I-BPNN) for forest fire prediction. Firstly, a probabilistic weighted rough set attribute reduction method is designed to effectively eliminate non-critical and redundant features in the dataset and simplify the input space of the neural network. Subsequently, a particle swarm optimization (PSO) algorithm is employed to refine the BP neural network (BPNN), aiming to elevate both the precision and efficiency of forest fire prediction. To validate the method’s effectiveness, experiments are conducted on three representative forest fire datasets. The results show that compared with the traditional machine learning prediction methods, the proposed forest fire prediction model achieves a significant improvement in prediction accuracy and is more suitable for early warning and disaster prevention and mitigation strategies in forest fire-prone areas.