A Hybrid Approach Combining LSTM Prediction, Genetic Algorithm Optimization, and Reinforcement Learning for Adaptive Traffic Signal Management
Abstract
The dynamically growing urban population and the number of vehicles have exacerbated traffic congestion; thus, intelligent and scalable traffic management systems have to be developed. This study will present an AI-based optimization system, which is based on the parallel computing, to be used in optimizing real-time traffic prediction, route, and signal control. The structure of the system consists of three layers linked to each other: the data collection layer, which processes traffic, weather, and accident data used by various sources; an AI processing layer, integrating models of deep learning and optimization models; and a decision layer reporting the results of the application to both traffic authorities and commuters. LSTM networks are deployed to predict traffic flow estimating the temporal relationships and forecasting congestion under different conditions. GA and RL are used in the optimization of routes and adaptive traffic signal, respectively, thereby means of efficient vehicle routes and shorter delays. The integration with distributed data processing engine (Apache Spark) and deep learning frameworks accelerated by GPU (TensorFlow / PyTorch) will be used to guarantee scalability and real-time performance. Validation was done experimentally using the SUMO simulation platform, which has proven effectiveness of the proposed framework. The results of the LSTM model demonstrated a much lower level of errors in the prediction (MAE = 0.23, RMSE = 0.28) than in its traditional counterparts. It was found that the RL-based signal controller yielded an average waiting time 28-percent lower per intersection and 18-percent higher throughput when compared to the fixed-time scheduling method. Moreover, parallel processing experiments indicated that the average data processing latency decreased when scaling Spark clusters between 4 and 16 nodes (reduction of 2.1s to 0.5s), which means it is robust and scalable.
References
Chen, M., & Zhou, Q. (2023). A Spark-based streaming framework for large-scale smart city data. The Journal of Supercomputing, 79(14), 15033–15048. https://doi.org/10.1007/s11227-023-05211-5
David, O. E., & Greental, I. (2014). Genetic algorithms for evolving deep neural networks. In Proceedings of the Companion Publication of the 2014 Annual Conference on Genetic and Evolutionary Computation (pp. 1451–1452). Association for Computing Machinery. https://doi.org/10.1145/2598394.2602287
Du, W., Ye, J., Gu, J., Li, J., Wei, H., & Wang, G. (2022). SafeLight: A reinforcement learning framework for collision-free traffic signal management. arXiv. https://arxiv.org/abs/2211.10871
Graves, A., & Jaitly, N. (2014). Towards end-to-end speech recognition with recurrent neural networks. In Proceedings of the 31st International Conference on Machine Learning (pp. 1764–1772). PMLR.
Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735
Hossain, D., & Capi, G. (2017). Genetic algorithm based deep learning parameters tuning for robot object recognition and grasping. Journal of Advanced Computational Intelligence and Intelligent Informatics, 21(2), 264–273. https://doi.org/10.20965/jaciii.2017.p0264
Huang, Y., & Chen, L. (2023). The role of weather in short-term traffic flow prediction. Sustainability, 15(6), Article 2576. https://doi.org/10.3390/su15062576
Kumar, J., Goomer, R., & Singh, A. K. (2018). Long short term memory recurrent neural network (LSTM-RNN) based workload forecasting model for cloud datacenters. Journal of Grid Computing, 16(4), 1–22. https://doi.org/10.1007/s10723-018-9457-3
Li, J., Wu, K., & Hu, Y. (2021). Cloud-based predictive traffic control using Spark parallel computing. IEEE Transactions on Intelligent Transportation Systems, 22(9), 5522–5534. https://doi.org/10.1109/TITS.2020.3012345
Lin, Y., Chen, H., & Wu, J. (2025). A convolutional LSTM model capturing deep spatio-temporal dependencies for traffic prediction. Scientific Reports, 15, Article 11234. https://doi.org/10.1038/s41598-025-11234-5
Nikravesh, A., Ajila, S., & Lung, C. H. (2016, July). Mobile network traffic prediction using MLP, MLPWD, and SVM [Paper presentation]. 2016 IEEE International Congress on Big Data (BigData Congress), San Francisco, CA, USA (pp. 402–409).
Park, D. C. (2009). Structure optimization of bilinear recurrent neural networks and its application to Ethernet network traffic prediction. Information Sciences, 179(8), 1152–1161. https://doi.org/10.1016/j.ins.2008.12.006
Rahman, M., & Karim, A. (2023). Performance evaluation of machine learning algorithms in SparkMLlib. Algorithms, 16(2), Article 74. https://doi.org/10.3390/a16020074
Specht, D. F. (1991). A general regression neural network. IEEE Transactions on Neural Networks, 2(6), 568–576. https://doi.org/10.1109/72.97934
Sutskever, I., Vinyals, O., & Le, Q. V. (2014). Sequence to sequence learning with neural networks. In Z. Ghahramani, M. Welling, C. Cortes, N. Lawrence, & K. Q. Weinberger (Eds.), Advances in Neural Information Processing Systems 27 (pp. 3104–3112). Curran Associates, Inc.
Vanli, N. D., Sayin, M. O., Delibalta, I., & Kozat, S. S. (2017). Sequential nonlinear learning for distributed multiagent systems via extreme learning machines. IEEE Transactions on Neural Networks and Learning Systems, 28(3), 480–492. https://doi.org/10.1109/TNNLS.2016.2517017
Wang, L., & Gao, Y. (2023). Methods for predicting traffic flow in intelligent transportation: A review. Applied Sciences, 13(7), Article 3866. https://doi.org/10.3390/app13073866
Xu, L., Cheng, X., Liu, Y., & Xu, B. (2015, October). Mobility load balancing aware radio resource allocation scheme for LTE-advanced cellular networks [Paper presentation]. 2015 International Symposium on Wireless Communication Systems (ISWCS), Brussels, Belgium (pp. 61–65).
Xu, L., Luan, Y., Cheng, X., & Chao, K. S. (2016, September). Self-optimised joint traffic offloading in heterogeneous cellular networks [Paper presentation]. 2016 IEEE 27th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC), Valencia, Spain (pp. 1–6).
ZareMoayedi, H., & Masnadi-Shirazi, M. A. (2008, September). ARIMA model for network traffic prediction and anomaly detection [Paper presentation]. 2008 International Symposium on Information Technology, Kuala Lumpur, Malaysia (Vol. 4, pp. 1–6).
Zhang, H., & Li, X. (2023). Reinforcement learning and predictive analytics for dynamic optimization of traffic flows. Sustainability, 15(8), Article 3383. https://doi.org/10.3390/su15083383
Zhang, X., Li, Q., & Wang, J. (2025). Federated deep reinforcement learning approach for managing urban traffic signal control. Scientific Reports, 15, Article 8732. https://doi.org/10.1038/s41598-025-08732-5
