MACHINE LEARNING FOR TEXT CLASSIFICATION IN BUILDING MANAGEMENT SYSTEMS

In building management systems (BMS), a medium may have between 200 and 1000 sensor points. Their labels need to be translated into naming standard so they can automatically recognised by the BMS platform. The current industrial practices often manually translate these points (this is known as tagging process), which takes around 8 hours for every 100 We introduce an AI-based multi-stage text classification that translates formatted labels. After comparing five different techniques (logistic regression, random forests, XGBoost, multinomial Naive Bayes linear support vector classification), we demonstrate XGBoost top performer with 90.29% of true positives, use prediction confidence filter out false positives. This approach applied in sensors networks various applications, where manual free-text data pre-processing remains cumbersome.