Work Disability Risk Prediction with Text Classification of Medical Reports

Abstract

Due to digitalization, more and more data on an individual’s well-being is available in various repositories owned by different organizations. Intelligent data processing methods such as machine learning, enable efficient and accurate value creation from data. This paper addresses the problem of how to process big and mostly unstructured data to predict the work disability risk of an individual. Currently, the best data for predicting disability risk of an individual comes from health and employment records. However, no simple indicator can be reliably used to detect the risk. In our work, we present a ML model for assessing the risk of losing work ability based on anonymized medical reports of an occupational health care company. Our models are created using the ULMFit toolset and they reach accuracy of 72 % in a two class case and 65% in a three class case.