Bus Arrival Time Prediction Using Hybrid LightGBM-LSTM

As a critical factor in urban traffic management and intelligent transportation systems, the reliability of bus arrival time prediction significantly impacts the efficiency of public transportation operations and passenger satisfaction. Nevertheless, real-world complexities such as variable traffic conditions, unexpected road incidents, and weather changes introduce considerable difficulties in forecasting precise arrival times. Traditional machine learning methods like SVM and Boosting algorithms have been used to predict bus arrival times. However, these models often fail to capture the time-dependent and sequential patterns inherent in time-series data, such as bus schedules. On the other hand, neural network models excel at handling sequential data and can effectively model temporal dependencies but may encounter challenges when dealing with large-scale complex datasets. Given the limitations of a single predictive model in dealing with the multifaceted nature of bus arrival times, this paper proposes a hybrid model to overcome the shortcomings of a single model in handling both long-term and short-term data patterns. The hybrid model aims to improve prediction accuracy by utilizing the efficiency of LightGBM in feature selection and ensemble learning, as well as the advantages of LSTM in temporal data analysis.