Text classification with machine learning: An example of a maintenance management system

Today, it is important for companies to manage information and to make processes independent from individuals. Information obtained by an employee should be able to be conveyed to other employees and the information should be permanent in the company even if the employee leaves the workplace. For this reason, the needs of companies for information management systems are increasing. It is necessary to use an electronic management program in order to manage all assets such as machinery, equipment, heating and cooling systems installed in companies, to organize maintenance processes in a healthy way. Asset and maintenance management programs written to meet this requirement provide the desired benefits only when they can be used effectively. This study focuses on the process of recording malfunctions in assets, which is one of the main functions of the asset and maintenance management program used in the yarn company. The recommended machine learning technique for facilitating and accelerating the registration process and reducing human error is explained. With this technique, it is aimed to classify the fault description written during the recording with machine learning and assign it to the relevant fault category. With the study, it is expected that the work of those who enter data will become easier and faster, those who will perform maintenance have access to the right information, and the more effective and efficient use of the system will improve the process. In order to classify the fault descriptions in the maintenance management system according to the fault definitions with machine learning, in this study, the attributes were extracted with bag of word methods and the model was trained with logistic regression, linear support vector classification (SVC), stochastic gradient descent - classifier (SGDC), Naive Bayes (NB) algorithms. With the Linear SVC algorithm and the TF-IDF feature extraction method, a model was created that can predict 46 fault definitions corresponding to 72.12% of all failures in the maintenance management system with 87% success.