Neuro-fuzzy system for PM-10 prediction in the Northern Region

Abstract

Each year during summer, open-air burning is a major issue that contributes to the air pollution problem in the northern region of Thailand. One of the five major air pollutions is particulate matter with an aerodynamic diameter smaller than 10 micrometers, called PMio, and this regularly exceeds the national air pollution standard. In order to improve the effectiveness of planning for future outdoor activities or creating of a notification system, a data mining technique is implemented to create a prediction model. Data from eight Pollution Control Department fixed-site data monitoring stations located in the northern region from 2010 to 2018 were collected to develop a model. This study proposes Neuro-Fuzzy Transformation with the Minimize Entropy Principle model that applies a novel process of data transformation. Firstly, input features are applied with the Minimize Entropy Principle approach to identify membership functions. Then, membership values are created and transformed to new input features as historical data by the Collective Neural Networks called C-ANN. The proposed model is used to predict PMro classes one to three hours ahead. To evaluate the models in this work, the accuracy and F-score of each class are utilized. The results of the prediction models in this study shows that average accuracies were 94.98, 84.86, and 74.87 percent, and average F-scores were 0.78, 0.74, and 0.65 for two, three, and five classes, respectively.