Public awareness about the level of air quality has gradually increased the air quality inside a vehicle cabin. Air quality inside a vehicle cabin is affected by several gases, namely, carbon dioxide (CO2), particulate matter (PM), nitrogen dioxide (NO2), sulphur dioxide (SO2), ozone (O3) and carbon monoxide (CO). Majority named gases are harmful to health. Most of the drivers and passengers usually shut the windows and switch the vehicle's ventilation system into recirculation mode. In such confined space, pollution source mainly comes from occupants, whom produce carbon dioxide as contaminant known as human bioeffluent. The high concentration of carbon dioxide reduces human cognitive ability, causing dizziness and fatigue. Dizziness and fatigue can increase the driver's risk of getting into car accident. Consequently, this situation bring danger to the occupants, and to other potential road users as well. Existing air quality monitoring system does not provide in-vehicle air quality index and real-time monitoring and prediction of in-vehicle air quality. Therefore, this study developed a system named in-vehicle air quality monitoring system (IV-AQMS) that has been used to collect the air quality level inside the vehicle cabin. The system also includes a cloud-based database that stores the real-time air quality data from the targeted vehicles. The data obtained from the system was then used to predict future air quality inside a vehicle cabin with the recirculation ventilation mode with the purpose of avoiding vehicle occupants to stay in poor air quality environment. A dedicated in-vehicle air quality monitoring system has been developed. By using different models of machine learning as well as deep learning, the future air quality level in the vehicle for five minutes, ten minutes and twenty minutes time slot have been predicted. The proposed Gated Recurrent Unit prediction model is found as the most suitable prediction model for this application because it has the highest coefficient of determination value (R2) of 0.97 of and low root mean squared error (RMSE) value of 2.54 compared with other artificial intelligence models.
