FRMHPCA: Design of an Efficient Feature Representation Model for Monitoring Health Parameters Via Correlation Analysis

Health parameter monitoring for real-time environments is a multidomain task that involves parameter sensing, pre-processing, filtering, feature extraction, variance-based selection, classification, and post-processing operations. A large number of machine-learning based models are proposed for performing these tasks, but most of these models are either highly complex, or have lower efficiency when applied to real-time clinical use cases. To overcome these issues, this text proposes design of an efficient feature representation model for monitoring health parameters via correlation analysis. The proposed model initially collects health monitoring information sets from multiple sensor sources. These sets are processed via a multidomain feature extraction unit that combines frequency analysis, entropy analysis, convolutional analysis, and detailcomponent analysis. The extracted features from collected on-field sensors are stored for training purposes. While new samples given to a correlation engine, which assists in identification of similarity indices between individual features, and their respective disease types. Based on this correlation, an auto regressive integrated moving average (ARIMA) model is activated, which assists in identification of presence ampersand severity levels of different disease types. Due to use of the multidomain features ampersand their correlation levels, the proposed model is able to evaluate types ampersand severity levels for heart diseases, blood pressure related issues, and kidney related issues. The model was tested on these diseases, and showcased an improved performance of 8.3% in terms of accuracy, 4.5% in terms of precision, and 6.4% in terms of recall levels. Thus, making it useful for a wide variety of realtime clinical use cases.