The prediction of air pollutant levels plays an essential role in regulatory plans focused on controlling and mitigating air pollutants, such as particulate matter (PM). Even when elevated pollution episodes may be rare, they recirculate within an area where more people may present significant adverse health consequences. Thus, even when they are difficult to predict, pollution peaks prediction of air pollution particulate matter (PM2.5) is a crucial problem to address. Several machine learning (ML) approaches have been used to predict a set of air pollutants using different combinations of predictor parameters. Unfortunately, they are still not enough to generate accurate predictions of extreme values. This paper proposes a new hybrid method that combines the unsupervised learning Self-Organizing Maps with the supervised multilayer perceptron. The proposed method is applied for the prediction of extreme values of PM2.5, using five-year pollution data obtained from nine weather stations located in the metropolitan area of Santiago, Chile. Simulation results show that the hybrid method improves the performance metrics when predicting extreme values of PM2.5.