Análise do comportamento e percepções do consumidor de água utilizando técnicas de aprendizagem de máquina

Abstract

The main factors influencing the increase in water consumption today are population growth and changes in consumption patterns, as a result of economic development. Thus, the present study sought to predict the behavior of water users with the aid of machine learning. The data were collected through an online form, answered by students and professors of civil engineering course at the Federal University of Ceará, in Fortaleza. The analysis process took place in two stages: first, an exploratory analysis of the responses obtained was carried out and a level of consciousness was created with three possible classifications; and then the data were used to develop four classification-type decision tree (AD) models to predict the correlation between respondents' attitudes. The results showed that, despite the study population being in contact with the university environment and with easy access to information, only 19% of the total users were classified as high level, while the most were classified as having a medium level of awareness of consumption. The AD analysis reveals that the most significant variables in determining the level of consciousness were: frequency in the use of gray water for non-potable purposes; bath duration less than five minutes; watering the plants carried out in the early morning or early evening; and rainwater harvesting. The models showed accuracy above 70%, proving to be effective in describing and predicting the users' level of consciousness. In this way, the investigation of user behavior should be explored, to outline goals and solutions to reverse the current scenario of water insecurity. Therefore, the change must come, mainly, from the academic public, promoters of science and examples for the general society.