Fast AI classification for analyzing construction accidents claims

Abstract

Safety has long been considered an important issue in the construction industry. One means of reducing accidents is to provide for heavy compensation. As per the common law system, precedents, once made, then become part of the legal system. Therefore, construction companies and legal firms have an interest in obtaining details of court cases relevant to the ones they are currently involved with. However, the cost of identifying relevant court cases can be excessive. Computer-based text classification, the process of classifying documents into predefined categories with regard to their content, is proposed in this paper as a way to speed up the procedure whereby court cases are identified as relevant to a particular claim for accident compensation. The data set used for this project consisted of 3000 sentences. The 'training split' was 90% training and 10% for testing. The results show that the precision of the system proposed here for classifying construction accident cases is 95.7% and that the recall is 95.7%. This demonstrates that the fastText-based classification employed can successfully classify papers as relevant to the acceptance or rejection of a compensation case at a fairly high rate of accuracy. This pilot research provides a practical example in order to showcase the possibility of utilising artificial intelligence, without human intervention, for document classification. Such a facility could reduce the time taken to identify relevant past cases, so saving human resources, and improving turn-round times.