Construction accident narrative classification: An evaluation of text mining techniques.

•Evaluated six machine learning algorithms for classifying accident narratives in accordance to 11 accident types.•Found that support vector machine (SVM) produced the best performance.•Across the 11 accident types, the average precision of the SVM was 0.73, average recall was 0.63, and average F1 score was 0.67.•Commonly mislabeled cases were evaluated using confusion matrix, and qualitative evaluation of mislabeled cases.•A set of 1,000 labelled accident narratives and more than 3,000 unlabeled narratives were made available publicly.