Estimating passenger experience from vehicle data: Preconditions for using machine learning

These days, there is a vast amount of data available from public transport vehicles. This data has potential to be used for estimating passenger experience and comfort. Automatic and continuous monitoring of parameters affecting user experience can help in, for example, designing and developing better vehicles, using the information for operator rewards or for training drivers. This paper looks at how machine learning can be applied for estimating (electric) bus passenger experience automatically by using commonly available vehicle and sensor data. The focus is on how to ensure that data is interpreted correctly. We compare data collected from electric buses with passenger survey results and passenger feedback to identify relevant data types. Along with promising parameters for evaluating passenger experience, we provide recommendations on what needs to be considered in collecting actual user experiences for validating causation and results.