Anomaly detection and oversampling approach for classifying imbalanced data using CLUBS technique in IoT healthcare data

Abstract

Multiple data streams from sensing devices in intelligent settings have improved life quality thanks to the internet of things (IoT). Anomalies and imbalanced data sources are unavoidable due to system complexity and IoT device rollout issues. An imbalanced dataset has more data for one group than another, which may influence the results. IoT data streams are unbalanced, making anomaly detection harder. Data mining and machine learning classification approaches perform poorly on imbalanced datasets in the current setup. To address this, the proposed system suggests an effective anomaly detection method and oversampling approach (ADO) to improve IoT's ability to identify abnormal behaviours in imbalanced data. After clustering of lower and upper boundary standardisation (CLUBS) detects anomaly samples, the ADO technique provides synthetic samples for minority classes. ADO lowers class region overlaps and enhances classification methods. The experimental results using an imbalanced dataset and three classification algorithms, namely K-nearest neighbour (KNN), random forest (RF), support vector machine (SVM) and multilayer perceptron (MLP), show that the ADO approach increases classification accuracy (0.64% for KNN, 4.27% for RF and 6.33% for SVM) by removing anomalies and oversampling data.