Prediction of construction accident outcomes based on an imbalanced dataset through integrated resampling techniques and machine learning methods

Abstract

PurposeCentral to the entire discipline of construction safety management is the concept of construction accidents. Although distinctive progress has been made in safety management applications over the last decades, construction industry still accounts for a considerable percentage of all workplace fatalities across the world. This study aims to predict occupational accident outcomes based on national data using machine learning (ML) methods coupled with several resampling strategies.Design/methodology/approachOccupational accident dataset recorded in Turkey was collected. To deal with the class imbalance issue between the number of nonfatal and fatal accidents, the dataset was pre-processed with random under-sampling (RUS), random over-sampling (ROS) and synthetic minority over-sampling technique (SMOTE). In addition, random forest (RF), Naïve Bayes (NB), K-Nearest neighbor (KNN) and artificial neural networks (ANNs) were employed as ML methods to predict accident outcomes.FindingsThe results highlighted that the RF outperformed other methods when the dataset was preprocessed with RUS. The permutation importance results obtained through the RF exhibited that the number of past accidents in the company, worker's age, material used, number of workers in the company, accident year, and time of the accident were the most significant attributes.Practical implicationsThe proposed framework can be used in construction sites on a monthly-basis to detect workers who have a high probability to experience fatal accidents, which can be a valuable decision-making input for safety professionals to reduce the number of fatal accidents.Social implicationsPractitioners and occupational health and safety (OHS) departments of construction firms can focus on the most important attributes identified by analysis results to enhance the workers' quality of life and well-being.Originality/valueThe literature on accident outcome predictions is limited in terms of dealing with imbalanced dataset through integrated resampling techniques and ML methods in the construction safety domain. A novel utilization plan was proposed and enhanced by the analysis results.

Application of the Synthetic Over-Sampling Method to Increase the Sensitivity of Algorithm Classification for Class Imbalance in Small Spatial Datasets

“…The performance of the ML classifier for the AUC value of KNN-KM-SMOTE (0.89), and KNN-BL-SMOTE (0.82) is superior to the AUC value of KNN-RUS 0.68 [25]. The AUC value of RF-ADASYN-SVM-SMOTE (0.79) decreased by 0.04 from the AUC value of 0.83 of RF-RUS [25].…”

“…The authors of [25] develop a sampling technique scenario of RAW, RUS, ROS, and SMOTE to overcome the imbalance road accidents dataset with hyperparameter optimization using two techniques, namely random hold-out, and 5-fold CV, which were tested on the ML classifier RF, NB, KNN, and ANN. The AUC values were 0.83, 0.68, 0.76, and 0.78, respectively, based on the RUS, RUS, ROS, and ROS sampling techniques.…”

“…The highest AUC value in the Borderline SMOTE method for the SMOTE algorithm variant shows that the highest AUC value in SVM is 80.6% and KNN at 82%. The NB method obtained the lowest values of 55 Based on the literature study in Section 2, the experimental results in Table 5 refer to the research development [25]. The suggestions given for future work are to observe and apply the performance of the Nominal and Continuous-SMOTE (SMOTE-NC) oversampling techniques, Borderline-SMOTE (BL-SMOTE), K-Means-SMOTE (KM-SMOTE), and SVM-SMOTE as an effort to improve the performance of the ML classifier by testing the accident dataset, and in this case study we apply it to the road traffic accident dataset.…”

“…The results of the research [23] use SMOTE, Borderline SMOTE, and SVM SMOTE methods. The research result in paper [24,25] uses SMOTE, random over-sampling (ROS), and random under-sampling (RUS) method. Meanwhile, SMOTE-over-sampling and random under-sampling methods [26], and balanced bagging [27] were used to overcome imbalanced datasets obtained in realtime (time/hour).…”

“…[37] to improve the robustness [38,4]. The methods for the single ML classifiers are widely used to determine the effectiveness of the proposed model, including LR, DT, RF, and AdaBoost [24]; RF, NB, KNN, and ANNs [25]; Bayes classifiers [28]; binary classifiers [26,27]; SVMs and Probabilistic Neural Network (PNN) [23]; Decision Tree (DT), NB, kNN, Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), LR, SGD Classifier, SVMs, SVM-linear (SVM-L), SVM- RBF (SVM-R), SVM-polynomial (SVM-P), and MLP [39].…”