Application of a Fuzzy ARTMAP Neural Network for Indoor Air Quality Prediction

Abstract

Indoor air quality monitoring is an important activity to ensure continued health and well-being of citizens living, studying, and working in indoor environments. This practice has been widely developed through the application of low-cost sensors that are able to measure gas concentrations, particulate matter, and other components such as humidity and temperature that affect indoor air quality. Additionally, machine learning algorithms have been applied in the interpretation of sampled environmental data to improve the performance of monitoring systems. This paper proposes the implementation of a fuzzy ARTMAP neural network, which employs the concepts of Adaptive Resonance Theory (ART), to compute the prediction of particulate matter sampled in a domestic bedroom environment. With the application of a specialized online training architecture, the fuzzy ARTMAP network can be a promising alternative to predict particulate matter time series data modeled in sliding windows, obtaining predictions 24-hour ahead with mean absolute error (MAE) ranging here from 0.26 to 7.65.