Data Mining of Occupational Accidents in an Automotive Manufacturing Company for the Identification of Root Causes and the Implementation of Preventive Strategies

Abstract

Introduction: Contemporary automotive manufacturing enterprises, like other industrial sectors, routinely collect large volumes of data pertaining to productivity, occupational safety, and health. The transformation of these raw data into actionable knowledge has become increasingly critical for organizational decision- making. This study applies data mining techniques to analyze occupational accidents resulting in lost workdays at an Iranian automotive manufacturing company from June 2014 to March 2025, aiming to identify the patterns and determinants of workplace incidents.

Material and Methods: Accident records from the specified period were systematically collated and preprocessed, then subjected to classification and association rule mining using established data mining algorithms. Classifier implementations and experiments were conducted within the Weka software environment. In addition to supervised classification, association rule mining algorithms (Apriori and Predictive Apriori) were employed to discover co-occurring attribute patterns and identify combinations of factors associated with elevated accident risks.

Results: Through comparative evaluation, the support vector machine (SVM) classifier demonstrated the best performance in predicting accident type, severity & causes, achieving an accuracy of 66.08% under cross-validation. However, the decline from a 100% training accuracy indicates the presence of overfitting in the models. Furthermore, Apriori and Predictive Apriori algorithms successfully extracted rules with high performance metrics, demonstrating confidence levels above 83% and predictive accuracies between 97.9% and 98%. Quantitative analysis revealed that the highest correlation with incidents was found in a combination of specific factors: low work experience (0 to 5 years), late shift hours (4 to 8 hours after the shift start), and employment as a body production operator. Secondary contributing variables included shift scheduling, proximate causes of the event, annual production rate, and employee remuneration.

Conclusion: The findings demonstrate that data mining techniques are effective tools for uncovering hidden patterns in occupational accidents within the automotive industry; nevertheless, the observed overfitting in classification models remains a serious limitation, highlighting the need for more balanced datasets. From a practical perspective, the identified association rules enable the design of specialized training programs, optimized rest schedules, and targeted safety interventions specifically for high-risk groups, such as low-experienced workers (0–5 years) operating in body production during the late hours of their shifts (4–8 hours).