Discovering and recognizing of imbalance human activity in healthcare monitoring using hybrid SMOTE Tomek technique and decision tree model

Human activity recognition model is vital and has been use in healthcare monitoring system. Bespoke multi-modal sensors were used such as accelerometer, gyroscope, GPS, temperature, pressure mat etc. Hence, the activities involved may varied resulted on class imbalance issue therefore, the model accuracy also degraded and may not provide the desired results in all aspects. Resampling method addressed as Synthetically Minority Oversampling Technique and Tomek Link (SMOTE Tomek) is proposed to balance the target classes. Moreover, many classification algorithms such as Logistic Regression (LR), Support Vector Machine (SVM) and Decision Tree (DT) were selected for the experiments on two datasets namely MARBLE that was publicly available and MARDA dataset. The classification accuracy achieved 98.36 with hybrid SMOTE Tomek on MARBLE dataset and 97.45 with for the MARDA dataset with total execution time 19.4ms and 42.6ms respectively. Consequently, the proposed model can be deployed in a healthcare system dashboard for effective monitoring and efficient decision making.

Actions (login required)

View Item