Machine learning application for support for automated control systems users

Abstract

The article presents the results of the analysis of determining the possibility of Machine Learning (ML) using for solving the problems of incident classification of users on the example of enterprise resource planning (ERP) systems of JSC Russian Railways and choosing a rational method for solving this problem. The presented problem is a special case of the Natural Language Processing Problem (NLP) with the use of neural networks. The article covers findings of classification of user queries: binary and multi class, based on semantic analysis. In 2018, 1.4 million incidents were reported from users of ERP systems, only 25% of them are suitable for ML applications. Resolving other incidents requires the participation of a support specialist. Implementation of the software will reduce the waiting time for incident processing by 64% of the average. This will increase staff productivity by 5-7%. This paper explicitly proves that using convolutional neural networks for user incident classification problem has high potential in JSC Russian Railways.