A neuro-fuzzy and deep learning framework for accurate public transport demand forecasting: Leveraging spatial and temporal factors

Efficient public transportation requires innovative planning and operational strategies. Accurate demand forecasting is crucial, as it is influenced by complex, non-linear interactions of various spatial and temporal factors. This study proposes a neuro-fuzzy inference and deep learning models to predict public transport demand in Mashhad's traffic zones for enhanced operational planning. The model's flexibility allows the integration of diverse temporal and spatial variables. Four Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Long Short-Term Memory (LSTM) models developed with two datasets were evaluated and compared to each other. Datasets one and two contained all possible variables without pre-judging their impact, encompassing daily and yearly horizons, respectively. Datasets three and four employed the identified influential variables from previous datasets using the Random Forest algorithm, leading to faster processing and reduced error. Five statistical coefficients including MSE (Mean Squared Error), BIAS, R² (Coefficient of Determination), WI (Willmott Index) and NSE (Nash-Sutcliffe Efficiency were presented to evaluate the performance of the proposed models. The results showed that the LSTM neural network model in the short-term daily scale (MSE = 0.0006, BIAS = 0.9308, R² = 0.9047, WI = 0.7591, NSE = 0.9047) and the ANFIS model in the long-term annual scale (MSE = 0.0024, BIAS = 0.0229, R² = 0.9415, WI = 0.9730, NSE = 0.8738) achieved superior performance in predicting demand for bus and rail systems in Mashhad. This research's forecasting models enable planners to estimate public transport demand under varying utilization levels of urban uses in Mashhad, offering insights for both daily and annual horizons across different traffic zones.