A Genetic Algorithm-based AutoML Approach for Large-scale Traffic Speed Prediction

With the continuous innovation of computer science as well as the big data acquisition technology, machine learning (ML), developed as a state-of-art framework, now has been comprehensively applied in speed prediction tasks. However, ML methods usually require intensive hyper-parameter tuning, which hinders the practical deployment of ML models. In view of this, this paper proposes an automated machine learning (AutoML) framework for speed prediction, which enables the prediction work to be accomplished in a much more timesaving and convenient way as well as in high prediction accuracy. The proposed framework utilizes the Genetic Algorithm (GA) following its four major procedures: Genome coding, Crossover, Mutation and Selection to automatically search for the optimal neural network architectures and hyperparameters. The proposed framework is examined on a real-world large-scale dataset in the city of Berlin, Germany. The experimental results demonstrate that the proposed method outperforms other benchmarking methods by a significant margin. Sensitivity analysis is also conducted to show the robustness of the proposed method. This study demonstrates the great penitential of using AutoML in traffic speed prediction and other related transportation applications.