Cluster-aware deep learning framework for bus arrival time prediction in Indian cities

In transit service operations, accurate and reliable prediction of bus arrival times is essential. The accuracy of travel time prediction depends on both the quality of input data and the prediction algorithm employed. However, most reported studies have selected model inputs either arbitrarily or based on their availability rather than their suitability. In addition, most studies have evaluated prediction methods under a specific traffic and operational environment, limiting the generalizability of their findings. This study addresses these limitations by evaluating three prediction models using data from three Indian cities: Delhi, Mysore, and Surat. In Delhi and Mysore, buses operate under mixed traffic conditions, with Delhi experiencing high travel time variability and Mysore relatively low variability. In contrast, Surat operates a bus rapid transit system with dedicated bus lanes, representing a distinct operational environment. To identify significant input features, temporal analysis was used to determine the temporal features, while the deep embedded clustering algorithm was employed to exploit the spatial variations in travel times by grouping road segments with similar spatio-temporal characteristics. Separate prediction models were then developed for each cluster using long short-term memory (LSTM), convolutional LSTM (ConvLSTM), and ConvLSTM with an attention mechanism. Their performance was assessed and compared to determine their suitability for cities with varying operational characteristics. The results demonstrate that the proposed framework improves prediction accuracy by reducing route heterogeneity and effectively capturing localized spatio-temporal patterns. The findings reveal that the proposed approach not only enhances model accuracy but also offers practical guidance on selecting appropriate prediction models based on city-specific traffic variability, contributing towards more scalable and reliable bus transit operations.