AI-based Prediction of Independent Construction Safety Outcomes from Universal Attributes (1908.05972v2)

Abstract: This paper significantly improves on, and finishes to validate, an approach proposed in previous research in which safety outcomes were predicted from attributes with machine learning. Like in the original study, we use NLP to extract fundamental attributes from raw incident reports and machine learning models are trained to predict safety outcomes. The outcomes predicted here are injury severity, injury type, body part impacted, and incident type. However, unlike in the original study, safety outcomes were not extracted via NLP but were provided by independent human annotations, eliminating any potential source of artificial correlation between predictors and predictands. Results show that attributes are still highly predictive, confirming the validity of the original approach. Other improvements brought by the current study include the use of (1) a much larger dataset featuring more than 90,000 reports, (2) two new models, XGBoost and linear SVM (Support Vector Machines), (3) model stacking, (4) a more straightforward experimental setup with more appropriate performance metrics, and (5) an analysis of per-category attribute importance scores. Finally, the injury severity outcome is well predicted, which was not the case in the original study. This is a significant advancement.

• The paper refines construction safety prediction by using independent human annotations to eliminate artificial bias in outcome assessment.
• It utilizes advanced machine learning models on over 90,000 incident reports to accurately forecast injury severity, type, body part, and incident category.
• The study confirms that NLP tools are effective across various construction domains, providing actionable insights for improved safety management.

AI-based Prediction of Independent Construction Safety Outcomes

The paper "AI-based Prediction of Independent Construction Safety Outcomes from Universal Attributes" presents a significant advancement in the field of construction safety research. It builds upon earlier work by Tixier et al. (2016), where machine learning was leveraged to predict construction injury outcomes using NLP to extract attributes from incident reports. This paper offers an improved methodology that addresses limitations from the previous research, notably eliminating potential artificial correlations between predictors and outcomes by utilizing safety outcomes provided by independent human annotations.

Overview of the Study

The authors introduce a comprehensive paper leveraging machine learning techniques to analyze a substantial dataset, comprising over 90,000 construction incident reports. The NLP tool developed by Tixier et al. (2016) is used to extract fundamental contextual attributes from these reports. Subsequently, machine learning models, including Random Forest (RF), XGBoost, and linear Support Vector Machine (SVM), were trained to predict four key safety outcomes: injury severity, injury type, body part impacted, and incident type.

Key improvements introduced in this paper include:

1. Independent Annotation: Safety outcomes were provided by independent experts, avoiding biases that could arise when both predictors and outcomes are derived from the same source.
2. Expanded Dataset: Utilizing a much larger dataset enhances the reliability and generalizability of the results.
3. Advanced Machine Learning Techniques: Incorporation of models such as XGBoost and linear SVM, and the application of model stacking, allow for a robust evaluation of prediction performance.
4. Enhanced Experimental Setup: More appropriate performance metrics and straightforward configurations are adopted, alongside a detailed analysis of attribute importance per category.

Research Findings

The paper reports high predictive performance across the different models, with comparable efficacy among RF, XGBoost, and linear SVM methods, significantly surpassing random baseline predictions. Notably, injury severity was accurately predicted by all models, addressing a gap identified in previous studies where severity was considered unpredictable from attributes alone.

The results confirm the robustness of the NLP tool across various industrial domains beyond its initial development setting, demonstrating its viability in diverse construction environments, including the oil and gas sector.

Implications

The validation of the NLP and machine learning approach offers valuable insights into construction safety management. Organizations are encouraged to leverage their injury report databases, employing this approach to gain actionable insights that can enhance safety planning and incident prediction. The attribute-based methodology aids in overcoming cognitive biases prevalent in hazard recognition, potentially improving empirical decision-making in site management.

Speculation on Future Developments

Future research should explore the possibility of predicting the occurrence of incidents, requiring access to datasets representing non-incident cases. As machine learning technology evolves, newer algorithms can be integrated within this framework to increase predictive accuracy and efficiency.

Additionally, investigation into the human-machine interaction during worksite planning could elucidate how such analytical tools might augment human judgments, optimizing safety protocols and hazard forecasting methodologies.

In conclusion, this paper represents a methodological refinement in the prediction of construction safety outcomes, providing a template for implementing machine learning insights in real-world scenarios that can substantially enhance industry safety standards.