Traffic accidents continue to inflict devastating human and economic losses on a global scale, claiming approximately 1.35 million lives annually. This paper proposes STAPS-AI, an advanced Smart Traffic Accident Prediction System that tightly integrates classical machine learning (ML) algorithms with contemporary artificial intelligence (AI) architectures — including Long Short-Term Memory (LSTM) recurrent networks, Convolutional Neural Networks (CNN), and a Transformer-based attention model — to achieve robust, real-time accident risk prediction and severity classification. The system ingests heterogeneous data streams encompassing real-time traffic flow, historical accident records, meteorological readings, road geometry, driver-behavior telemetry, and connected-vehicle V2X communications, and fuses them within a unified preprocessing and feature-engineering pipeline. A comprehensive comparative evaluation is conducted across eight models — Logistic Regression, SVM, Random Forest, XGBoost, Multilayer Perceptron (MLP), LSTM, CNN-LSTM Hybrid, and a Transformer — on the UK Department for Transport Road Safety dataset comprising 1.83 million records (2015–2022). The Transformer model achieves the highest predictive performance with 96.1% accuracy, 95.8% precision, 95.2% recall, and an AUC-ROC of 0.983. A cost-sensitive four-tier severity classifier (Low / Moderate / High / Critical) provides granular, actionable risk intelligence directly consumable by traffic management centers, navigation platforms, and autonomous vehicle onboard systems. STAPS-AI outperforms all prior state-of-the-art baselines by margins of 4.8–17% in accuracy, and its modular REST-API architecture supports real-time deployment with median inference latency of 11 ms per 1,000 road
segments.
This paper presented STAPS-AI, an advanced Smart Traffic Accident Prediction System that integrates classical machine learning with state-of-the-art artificial intelligence to deliver real-time, multi-tier accident risk prediction at scale. Evaluated on
1.83 million real-world UK road accident records, the proposed Transformer-based model achieved 96.1% accuracy and an AUC-ROC of 0.983 — surpassing all prior baselines by at least 4.8 percentage points. The four-tier severity classification module provides operational risk intelligence across Low, Moderate, High, and Critical tiers, directly actionable by traffic management centres, emergency services, and autonomous vehicles. STAPS-AI’s modular REST-API architecture enables real-time deployment at 11 ms inference latency, meeting the stringent timing requirements of live ITS operations.
Future research will pursue three directions: (1) federated learning to enable privacy-preserving model training across multiple national road authorities without data centralisation; (2) incorporation of ADAS-derived driver-state signals to augment the prediction pipeline with real-time behavioural context; and (3) full end-to-end V2X integration to deliver per-vehicle onboard risk scores in sub-100 ms, enabling proactive collision avoidance within autonomous driving stacks.
References
1World Health Organization, “Global Status Report on Road Safety 2023,” WHO Press, Geneva, Switzerland, 2023.
2A. P. Tarko, “Modeling accident frequencies as zero-altered probability processes: an empirical inquiry,” Accident Analysis &
Prevention, vol. 43, no. 4, pp. 1517–1524, Jul. 2011.
3M. A. Abdel-Aty and A. E. Radwan, “Modeling traffic accident occurrence and involvement,” Accident Analysis & Prevention, vol. 32,
no. 5, pp. 633–642, Sep. 2000.
4S. P. Miaou and H. Lum, “Modeling vehicle accidents and highway geometric design relationships,” Accident Analysis & Prevention,
vol. 25, no. 6, pp. 689–709, Dec. 1993.
5J. Yuan, M. Abdel-Aty, and L. Qin, “Random forest-based real-time crash risk prediction on arterials using connected vehicle data,”
Accident Analysis & Prevention, vol. 135, p. 105371, Feb. 2020.
6A. Sheykhfard and F. Haghighi, “Performance analysis of four different algorithms for traffic accident classification under imbalanced
data,” Int. J. Injury Control and Safety Promotion, vol. 27, no. 4, pp. 403–413, 2020.
7T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” in Proc. 22nd ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining, San Francisco, CA, USA, Aug. 2016, pp. 785–794.
8K. Zhang, M. Hassan, and S. Yahaya, “Predicting traffic accidents using deep learning: an LSTM-based approach,” IEEE Trans. Intell. Transp. Syst., vol. 22, no. 5, pp. 2818–2829, May 2021.
9H. Yu, X. Liu, and Y. Chen, “Spatiotemporal accident prediction via convolutional neural networks on road grid maps,” Transp. Res.
Part C: Emerg. Technol., vol. 103, pp. 106–122, Jun. 2019.
10R. Moradi and A. Vagnoli, “Transformer-based accident prediction using vehicle telematics streams,” in Proc. IEEE Intelligent Vehicles
Symp. (IV), Anchorage, AK, USA, Jun. 2023, pp. 1015–1022.
11A. Vaswani et al., “Attention is all you need,” in Proc. Advances in Neural Information Processing Systems (NeurIPS), vol. 30, Long
Beach, CA, USA, Dec. 2017.
12S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, Nov. 1997.
13N. V. Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “SMOTE: Synthetic minority over-sampling technique,” J. Artificial Intelligence Research, vol. 16, pp. 321–357, Jun. 2002.
14M. Abdel-Aty, J. Huang, A. Kinoshita-Raby, and S. Das, “A real-time crash prediction framework using attention-based deep learning
for freeway operations,” IEEE Trans. Intell. Transp. Syst., vol. 24, no. 3, pp. 3143–3157, Mar. 2023.
15UK Department for Transport, “Road Safety Data 2015–2022,” Open Government Licence v3.0, available:
https://www.data.gov.uk/dataset/road-accidents-safety-data, accessed May 2026.
How to Cite This Paper
SADDIKOOTI BHANU PRAKASH REDDY, POTTURU JAYA KRISHANA, DENDUKURI PRAVEEN VARMA, Ms. M. Vaishnavi (2026). Smart Traffic Accident Prediction System Using Machine Learning and Artificial Intelligence. International Journal of Computer Techniques, 13(3). ISSN: 2394-2231.
