Machine Learning for Traffic Flow Prediction and Optimization

Abstract

This survey examines the integration of Explainable Artificial Intelligence (XAI) with Deep Learning (DL) for brain tumor detection and analysis using MRI images. While Convolutional Neural Networks (CNNs) achieve high accuracy in identifying brain tumors, their decision-making process remains largely opaque, limiting trust and adoption in clinical settings. XAI addresses this issue by making AI predictions interpretable through techniques such as Grad-CAM, Layer-wise Relevance Propagation (LRP), and SHAP, which highlight important image regions, pixels, or features influencing model decisions. The survey reviews recent research on brain tumor detection, including transfer learning, CNN-based classification, segmentation approaches, hybrid CNN-SVM models, and various XAI methods. Each technique offers unique advantages but also has limitations. Grad-CAM provides region-level visual explanations but lacks fine detail, LRP delivers pixel-level attribution but can be noisy, and SHAP offers feature-level insights at the cost of high computational complexity. Studies consistently show that no single XAI method provides a complete explanation of model behavior. To overcome these limitations, the survey advocates a unified XAI framework that combines multiple explainability techniques. In such a framework, Grad-CAM identifies suspicious tumor regions, LRP highlights important pixels, and SHAP explains feature contributions simultaneously. This multi-resolution explanation approach improves transparency, clinician confidence, and trustworthiness of AI-assisted diagnosis. The paper also discusses key challenges in deploying XAI-enhanced medical imaging systems, including the lack of standardized evaluation metrics for explanations, high computational costs, difficulties in real-time deployment, limited clinician-friendly interfaces, robustness concerns, and regulatory requirements. Future research directions include lightweight explainability methods, multimodal and longitudinal analysis, uncertainty estimation, privacy-preserving AI, and explainable foundation models for medical imaging.

Introduction

This study focuses on traffic flow prediction and optimization using machine learning and deep learning techniques. Previous research has shown that statistical models such as SARIMA and machine learning models like ANN, Random Forest, SVM, ANFIS, and LSTM can effectively predict traffic conditions. Among these, LSTM networks consistently achieve the highest accuracy by capturing long-term temporal patterns and incorporating external factors such as weather and road events. Hybrid approaches combining LSTM with SARIMA, BiGRU, attention mechanisms, and bidirectional architectures further improve prediction performance. However, existing studies face challenges including poor transferability across cities, high computational costs, limited integration with traffic optimization systems, and reliance on simulated rather than real IoT data.

To address these limitations, the proposed system collects traffic data from IoT sensors, GPS/navigation feeds, CCTV cameras, and the Bangalore Traffic Dataset (16,705 records). Data preprocessing includes cleaning, handling missing values, encoding categorical variables, and normalization. Key features such as vehicle density, average speed, lane occupancy, weather conditions, signal status, and historical traffic volumes are extracted for model training.

Five models are evaluated: Linear Regression, Decision Tree, Random Forest, SVM, and LSTM. The LSTM model consists of two layers (128 and 64 units) with dropout regularization and is trained to capture both short-term and long-term traffic patterns. Predicted congestion levels are then integrated into a traffic signal optimization module, which dynamically adjusts signal timings using a modified Webster’s formula. Additionally, Dijkstra’s algorithm is used for route optimization by identifying the least congested alternative routes.

Experimental results show that the LSTM model outperforms all other methods, achieving:

• 94.3% accuracy
• MAE: 1.93
• RMSE: 2.87
• R²: 0.99

The system also demonstrates strong scalability, maintaining performance as the number of sensor nodes increases from 50 to 200. It remains robust under missing data conditions, with only a small increase in prediction error even when 30% of sensor data is unavailable.

Traffic optimization simulations using SUMO reveal significant benefits:

• 22–31% reduction in intersection delays
• 18–26% reduction in vehicle queue lengths
• 14–19% reduction in average travel times
• 12–17% faster alternative route recommendations

Furthermore, integration with a cloud-based smart city dashboard provides real-time congestion monitoring, hotspot prediction, signal control recommendations, and emergency vehicle routing. The addition of YOLOv11-based incident detection reduces emergency response times by approximately 18%.

Conclusion

This paper presents a comprehensive ML based framework for traffic flow prediction and optimization tailored to Smart City environments. By integrating multi-source real-time and historical data with comparative evaluation of five ML algorithms, the proposed LSTM-based model achieves 94.3% prediction accuracy with an R² of 0.99 the highest among all tested models. The system\'s adaptive signal optimization module reduces intersection delays by up to 31% and average travel times by up to 19%, demonstrating significant practical impact over traditional fixed-time systems. The framework\'s scalability (only 8.6% MAE increase from 50 to 200 sensors) and robustness (only 15.9% MAE degradation at 30% data loss) confirm its readiness for large-scale urban deployment. Integration with Smart City infrastructure IoT sensors, cloud dashboards, navigation apps, and incident detection positions the system as a holistic solution for modern intelligent transportation challenges. Future work will extend the system to city-wide multi-intersection coordination, incorporate federated learning for data privacy, and develop a real-world pilot deployment at selected Bengaluru intersections.

References

[1] Abduljabbar, R., Dia, H., & Liyanage, S. (2025). Machine Learning Traffic Flow Prediction Models for Smart and Sustainable Traffic Management. Infrastructures, 10(7), 155. [2] Chimmwal, A. (2022). Deep learning-based traffic flow prediction using LSTM networks incorporating weather and road events. Journal of Urban Transportation Systems, 14(3), 221–235. [3] Gandhi, P.& Ghosh, S. (2019). Unsupervised Twitter-based traffic data classification using word2vec semantic similarity. Indian Journal of Computer Science, 8(2), 94–108. [4] Jayachandra, R. (2024). Multi-model traffic flow forecasting: Integrating LSTM and SARIMA for real-world junction data. Transportation Letters, 16(4), 310–323. [5] Jha, A. (2025). ML framework integrating geospatial data and ANN for real-time traffic congestion hotspot prediction. IEEE Transactions on Intelligent Transportation Systems, 26(1), 45–57. [6] Kumar, A. et al. (2024). Navigating urban mobility: A review of AI-driven traffic flow management in smart cities. Proceedings ICRITO 2024, IEEE. [7] Kumar, S., & Vanajakshi, L. (2015). Short-term traffic flow prediction using Seasonal ARIMA model with limited input data. European Transport Research Review, 7(3), 1–9. [8] Li, C., & Xu, P. (2021). Application on traffic flow prediction of machine learning in intelligent transportation. Neural Computing and Applications, 33(2), 613–624. [9] Rajabhushanam, C. (2021). Comparative evaluation of DAE, DBN, RF, and LSTM for urban traffic flow prediction. International Journal of Advanced Research in Computer Science, 12(4), 88–96. [10] Rajalakshmi, V., & Vaidyanathan, R. (2021). Hybrid ARIMA-MLP and ARIMA-RNN models for traffic flow time-series forecasting. Applied Soft Computing, 103, 107159. [11] Sathyanarayana, K. (2023). Comprehensive assessment of deep learning methods for traffic prediction in intelligent transportation systems. IEEE Access, 11, 45321–45340. [12] Skoropad, V. N. et al. (2025). Dynamic Traffic Flow Optimization Using Reinforcement Learning and Predictive Analytics. Sustainability, 17(8), 3383. [13] Srinivasan, A., & Rupakesavan, G. (2024). EDA-LSTM framework for urban traffic volume forecasting. Transport Policy, 148, 77–88. [14] Tripathi, S., & Sharma, R. (2024). ANFIS-based traffic volume prediction with 89.91% accuracy for multi-vehicle-category datasets. Journal of Traffic and Transportation Engineering, 11(2), 233–245. [15] Vishnoi, P., & Kapoor, R. (2020). Traffic flow analysis and signal timing optimization using Webster\'s formula at J.L.N. Marg, Jaipur. International Journal of Civil Engineering, 7(5), 44–52. [16] Xu, Q. et al. (2025). Hybrid LSTM-BiGRU Attention model for urban traffic prediction using multi-resolution time windows. Transportation Research Part C, 162, 104612. [17] Yilin Han (2025). Scalable spatio-temporal graph neural network for urban traffic flow prediction. IEEE Transactions on Intelligent Transportation Systems, 26(3), 1123–1135.

Copyright

Copyright © 2026 Amarsinh B. Landage, Dheerajsing S. Thakur, Ajit S. Dhanaji, Vinod D. Shinde. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.