Harnessing the Multimodal Data Integration and Deep Learning for Occupational Injury Severity Prediction

Most previous studies have neglected the potential of integrating structured data and unstructured workplace injury reports to perform a predictive analysis of occupational injury severity. This study proposes an optimized integrated approach for occupational injury severity prediction using multimodal machine and deep learning techniques. We used 66,405 data points gathered from the US OSHA Severe Injury Reports from January 2015 to July 2021. Structured labeled data are preprocessed and normalized, whereas unstructured injury reports undergo text cleaning using Natural Language Processing techniques and text representation using Term Frequency-Inverse Document Frequency (TF-IDF) and Global Vector (GloVe) to convert them into numerical representations. Both modalities, in the form of vector representations, were concatenated and fed as input features for the proposed models. Seven sets of classifiers, namely Naïve Bayes, Support Vector Machine, Random Forest, Decision Tree, K-Nearest Neighbors, Long Short-Term Memory, and Bidirectional Long Short-Term Memory, were employed to learn the multimodal representations. The algorithm with superior performance was further optimized using the proposed feature importance and hyperparameter optimization techniques. Our findings revealed that the proposed optimized-Bi-LSTM architecture outperformed other classifiers in learning multimodal data to predict the likelihood of hospitalization and amputation with higher accuracies of 0.93 and 0.99, respectively. Consequently, the proposed approach enhances the performance by significantly improving the model processing time. This performance prediction provides a convincing benchmark for the successful execution of multimodal deep learning in occupational injury research. Therefore, the proposed multimodal occupational injury severity prediction model enhances the early screening and identification of at-risk workers with severe occupational injury outcomes, as well as, provides valuable information to improve the workplace safety, health, and well-being of the workers.