Fuel Efficient Routes Using Vehicular Sensor Data

This paper presents an application of Deep Neural Networks to vehicular sensor data. The first goal of this work is to produce artificial readings for two sensors that have missing values: CO2 and fuel consumption. A neural network was trained to produce these values, and it is able to capture the rough behavior of these two sensors, although it misses minor variations. To train a Multilayer Perceptron (MLP), we used data from the enviroCar project, which collects sensor readings from volunteers around the world. With the dataset containing values generated by the MLP, we investigated the effect of these observations in tracing routes focused on fuel efficiency over a graph based on the traffic network of Monchengladbach. Results show that the imputed values increase the estimated fuel consumption in an average of 15% and CO2 in 17% for all routes.