AUTOMATED MACHINE LEARNING FOR HUMIDITY, TEMPERATURE AND LIGHTING PREDICTION IN INDOOR ENVIRONMENTS

Abstract

Some environmental factors affect human’s health and well-being and can be harmful when its values are outside the ideal range. By using Internet of Things (IoT) devices, it is possible to measure indicators of these factors, enabling a better analysis of indoor environmental quality. With Machine Learning (ML), it is possible to analyze big data in a fast and accurate way, making it possible to create models that can predict future values based on past values recorded in a database. This study seeks to use Automated Machine Learning (Auto ML) to get the most effective machine learning predictive model for humidity, temperature, and lighting in indoor environments. The data used for the experiment were collected from two Smart Citizen Kits, located at two different indoor environments of an educational institution in Brazil. A comparison was made between different learning hypotheses and predictive models in order to find the most effective one, considering accuracy and speed. The results showed that the predictive models that used the Voting Ensemble algorithm along with XGBoostRegressor and LightGBM were the most effective ones, achieving an accuracy similar to the models that used Stack Ensemble, but in a considerably shorter time. This study will contribute to future researchers who need to analyze big data involving some of the variables analyzed in this study, as well as to help in the process of finding a good machine learning predictive model for other variables by using Auto ML.