Estimation of the fuel consumption in autonomous vehicles using machine learning techniques

Energy consumption is the responsibility of each individual and that we must all adopt the right reflexes to preserve our resources. So many reasons that can impact our fuel consumption such as nervous driving, under-inflated tires, poorly maintained vehicle…etc. Adopting economical and responsible driving can allow us to reduce our fuel budget while reducing our carbon footprint. The main objective of this research is to study the influence of some parameters on fuel consumption through the use of artificial intelligence approach. This paper presents the application of five machine learning techniques to estimate fuel consumption of the light cars, in particular, Linear Regression (LR), Random Forest (RF), Decision Tree (DT), K-Nearest Neighbors (KNN), Artificial Neural Network (ANN) and finally Deep Neural Network. These models have been developed using a dataset contains lot of criteria such us speed, distance and some weather conditions (rain, sun…). Performance of the algorithms is evaluated by reporting the prediction errors using the three metrics tools, which are median absolute error, mean absolute error and mean squared error. The statistical evaluation procedure finds that ANNs have the lowest median absolute error and mean squared error (0.304 and 0.357) compared to LR (0.401 and 0.919), RF(0.330 and 0.366), DT(0.400 and 1.177) and KNN(0.340 and 0.493) and the DNN have lowest mean absolute error (0.408) and mean squared error (0.305).