A collaborative privacy-preserving approach for passenger demand forecasting of autonomous taxis empowered by federated learning in smart cities

The concept of Autonomous Taxis (ATs) has witnessed a remarkable surge in popularity in recent years, paving the way toward future smart cities. However, accurately forecasting passenger demand for ATs remains a significant challenge. Traditional approaches for passenger demand forecasting often rely on centralized data collection and analysis, which can raise privacy concerns and incur high communication costs. To address these challenges, We propose a collaborative model using Federated Learning (FL) for passenger demand forecasting in smart city transportation systems. Our proposed approach enables ATs in different regions of the smart city to collaboratively learn and improve their demand forecasting models through FL while preserving the privacy of passenger data. We use several backpropagation neural networks as local models for collaborating to train the global model without directly sharing their data. The local model shares only the model updates with a global model that aggregates them, which is then sent back to local models to improve them. Our collaborative approach reduces privacy concerns and communication costs by facilitating learning from each other's data without direct data sharing. We evaluate our approach using a real-world dataset of over 4500 taxis in Bangkok, Thailand. By utilizing MATLAB2022b, the proposed approach is compared with popular baseline methods and existing research on taxi demand forecasting systems. Results demonstrate that our proposed approach outperforms in passenger demand forecasting, surpassing existing methods in terms of model accuracy, privacy preservation, and performance metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and R-squared (Formula: see text). Furthermore, our approach exhibits improved performance over time through the collaborative learning process as more data becomes available.