Predicting Vehicle Behavior Using Multi-task Ensemble Learning

Vehicle utilization analysis is an essential tool for manufacturers to understand customer needs, improve equipment uptime, and collect information future vehicle service development. Typically today, this behavioral modeling done on high-resolution time-resolved data with features such as GPS position fuel consumption. However, costly transfer sensitive from a privacy perspective. Therefore, typically only collected when the pays extra services relying that data. This motivated us develop multi-task ensemble approach knowledge enable behavior prediction low-resolution but high dimensional aggregated over time in vehicles. study proposes snapshot-stacked (MTSSE) deep neural network by considering vehicles’ operational life records. The utilizes measurements map low-frequency usage behaviors defined Two sources are integrated used: called Dynafleet, so-called Logged Data (LVD). experimental results demonstrate proposed approach’s effectiveness predicting low frequency With suggested network, it shown how sensor can highly contribute multiple simultaneously while using one single training process. • A snapshot-ensemble proposed. relies predict long term behavior. positively negatively affect performance. Extract task transference find tasks should be trained together.