Research Article
Accident Severity Prediction Using Hybrid Stacking of Machine Learning Models
@INPROCEEDINGS{10.4108/eai.28-4-2025.2358021, author={Pavani Penke and Lakshmi Srija Gandepalli and Kannam Venkata Pravallika and Chittiboyina Mounika Padma Sai and Buddaraju Amar Siva and Malipeddi. N.V.G.A. Deepthi}, title={Accident Severity Prediction Using Hybrid Stacking of Machine Learning Models}, proceedings={Proceedings of the 4th International Conference on Information Technology, Civil Innovation, Science, and Management, ICITSM 2025, 28-29 April 2025, Tiruchengode, Tamil Nadu, India, Part II}, publisher={EAI}, proceedings_a={ICITSM PART II}, year={2025}, month={10}, keywords={accident prediction road safety hybrid machine learning stacking ensemble xgboost catboost random forest logistic regression traffic density weather conditions feature engineering model evaluation}, doi={10.4108/eai.28-4-2025.2358021} }- Pavani Penke
Lakshmi Srija Gandepalli
Kannam Venkata Pravallika
Chittiboyina Mounika Padma Sai
Buddaraju Amar Siva
Malipeddi. N.V.G.A. Deepthi
Year: 2025
Accident Severity Prediction Using Hybrid Stacking of Machine Learning Models
ICITSM PART II
EAI
DOI: 10.4108/eai.28-4-2025.2358021
Abstract
Prediction of road accidents is important for road safety and decreasing the road crashes. This proposal aims to establish a robust hybrid machine learning model for predicting accidents by taking elements including weather, type of road, hour of day, traffic flow and driver as input factors. The proposed model uses a stacking ensemble learning method by aggregating various base model, including XGBoost, CatBoost, and Random Forest model combined with Logistic Regression as the meta-model, to increase pre- diction accuracy. The project begins by heavy pre-processing of data to ensure a good input for our models. Feature engineering methods are used to generate new features that can improve the ability of the model to make predictions. The performance of the hybrid model on testing is tested after the model is trained and compared with that of individual base models in terms of parameters, such as accuracy. The results demonstrate that the stacking model attains a high degree of accuracy, surpassing 95%, indicating its effectiveness for predicting accidents. This approach holds significant potential for enhancing safety measures on roadways and contributing to data-driven decision-making in traffic management.
