An Enhanced Framework for Anomaly Detection in Wireless Body Area Networks Using Gaussian Mixture Models and Random Forests

Wireless Body Area Network (WBAN) is a technology that uses sensors placed over different parts of the body to measure metrics necessary for monitoring a person's health. This technology finds its significance mainly in medical and healthcare systems, and enhances remote health monitoring of the patient. The sensors are mounted on the patient's body or are embedded into the body based on the desired measurements to be made. This technology helps save lives by monitoring the state of the body and diagnosing diseases from the metrics measured by the sensors. Being wireless in nature, sensors have a good potential to measure inaccurate information due to hardware failures or battery depletion. This reduces the confidence in the data measured by these sensors. Our research aims at detecting these anomalous data and classifying the anomalous data from normal data. This paper presents a combination of Gaussian Mixture Model (GMM) and Random Forest Classifiers (RF) to detect anomalies. Initially, the model performs preprocessing steps like feature selection, standardization and feature augmentation. The Gaussian Mixture Model is then fitted to the training data. The class imbalance is handled using SMOTE (Synthetic Minority Over-Sampling Technique). The model is then trained using Random Forest Classifier and is evaluated on test data set.