GAs-Based Weight Optimization of Multilayer Perceptron Neural Networks for Air Quality Prediction

Air pollution, particularly the concentration of particulate matter (PM10), poses significant risks to public health and environmental quality.Therefore, this study proposes a Multilayer Perceptron (MLP) optimized by a Genetic Algorithm (GA) to develop a predictive modeling approach for estimating PM 10 levels, capturing the complex interactions between atmospheric conditions and pollutants.The model incorporates twelve key input variables, including meteorological conditions and pollutant indicators such as temperature, CO, NO, NO 2 , NO x , PM 2.5 , O 3 , RH, SO 2 , wind direction (WD), wind speed (WS), and lagged PM 10 values.The dataset was divided into 70% for training and 30% for testing.The recommended model demonstrated a remarkable capacity to identify complex patterns in the PM 10 data by generating remarkably accurate predictions and a strong correlation with actual results during training.These results demonstrate the effectiveness of evolutionary optimization in enhancing FNN-based models for predicting air quality.