An Optimal Approach to Vehicular CO2 Emissions Prediction using Deep Learning

Abstract

One of the biggest challenges faced by humanity today is climate change. Governmental Organisations and Au-thorities all across the world, are now taking important steps to tackle this hazard, which if not dealt with, has potential of causing severe catastrophical damage, including the extinction of entire human species. One of the major contributors to this phenomenon is emissions from transport or vehicular emissions, which contribute significantly to the atmospheric concentration of CO2 or carbon dioxide, a greenhouse gas majorly responsible for climate change. The use of expensive and specialized sensors to monitor CO2 emissions in vehicles can be done, but it is neither scalable nor effective. In the proposed work, we suggest a feasible, efficient and scalable system to monitor these emissions, wherein the system proposed could be deployed on cloud, and receive the input sensor readings via IoT based dongles installed at the vehicular end, and predict the CO2 emissions. A 2-layer Long Short Term Memory (LSTM) network has been used in the proposed model, which is trained and tested on publicly available On-Board Diagnostics-II (OBD-II) data, and is compared with existing state-of-the-art model.