Prediction of monthly vehicle valorization/devaluation in Brazil with a MultiLayer Perceptron Regressor: a case study based on past sales, inflation, and interest rate

This work presents a comparison between the valuation/depreciation prediction results (from one month to another) of vehicles in Brazil considering the combination of four groups of characteristics: i) previous sales; ii) the number of vehicle sales; iii) basic interest rate; and iv) national consumer price index. We create a comparison baseline training a MultiLayer Perceptron Regressor (MLPR) based only on the vehicle’s value in the previous month, and then we train the MLPR by combining the previous vehicle value with combinations of the characteristic groups. Experiments were performed from 2013 to 2022 and evaluated in terms of Mean Squared Error (MSR) and Median Absolute Error (MAE). The combination of characteristics that presented the best MSR for the 2018-2022 period (COVID-19 period) was among the worst from 2014 to 2017. It is possibly concluded that data scientists must periodically adjust parameters according to the current economic conditions to obtain the best automatic forecast results of the monthly valorization/depreciation of vehicles in Brazil.