Design of a Highly Accurate PPG Sensing Interface via Multimodal Ensemble Classification Architecture

Abstract

Photoplethysmogram (PPG) sensing is a field of signal measurement that involves accurate sensor design and efficient signal processing. Sensing interfaces have matured due to use of sophisticated nano-meter technologies, that allow for high speed, and low error sampling. Thus, in order to improve the efficiency of PPG sensing, the signal processing unit must be tweaked. A wide variety of algorithms have been proposed by researchers that use different classification models for signal conditioning and error reduction. When applied to blood pressure (BP) monitoring, the efficiency of these models is limited by their ability to differentiate between BP levels. In order to improve this efficiency, the underlying text proposes a novel multimodal ensemble classifier. The proposed classifier accumulates correct classification instances from a series of highly efficient classifiers in order to enhance the efficiency of PPG sensing. This efficiency is compared with standard classification models like k-nearest neighbors (kNN), random forest (RF), linear support vector machine (LSVM), multilayer perceptron (MLP), and logistic regression (LR). It is observed that the proposed model is 10% efficient than these models in terms of classification accuracy; and thus, can be used for real time BP monitoring PPG signal acquisition scenarios. This accuracy is estimated by comparing actual BP values with measured BP values, and then evaluating error difference w.r.t. other algorithms.