Employer liability in work accident: A machine learning application from the health and safety at work perspective in Colombia

Abstract

Background

Work accidents have been studied from the perspective of outcome indicators, quantifying the frequency and severity of the occurrences. However, accidents at work continue to be a problem. In 2019, 611,275 accidents occurred in Colombia. All of this led workers to file lawsuits against employers, resulting in rulings by the Supreme Court of Justice's Labor Cassation Chamber—a topic that has been little studied.

Objective

This paper aims to develop and evaluate a machine learning model, the Surface Tension Neural Net-Decision Tree (STNN-DT), for classifying and predicting the favorability of verdicts related to employer liability in work accidents in Colombia.

Methods

This cross-sectional study defines and trains the proposed system based on data collected from 76 different cases in the Colombian context between 2019 to 2020. These cases correspond to actual verdicts issued by the Labor Cassation Chamber of the Colombian Supreme Court of Justice. An equal proportion of cases ruled in favor of and against employers was maintained. Methodologically, the study applied machine learning algorithms; a Bayesian Network and Surface Tension Neural Network-Decision Tree (STNN-DT).

Results

The results show that, in 85.43% of cases, the classifier Surface Tension Neural Network-Decision Tree correctly identified the judgments due to employer liability.

Conclusions

A striking 97% of the sentences for non-material damage reveal non-compliance by employers with legal regulations that predate current Safety and Health at Work standards. These violations primarily occur in IV-V risk companies, including mining, construction, and the installation of electrical networks.