Wearable Bio-Signal(PPG)-Based Personal Authentication Method Using Random Forest and Period Setting Considering the Feature of PPG Signals

摘要

A study regarding personal authentication based on PPG has been conducted using the Random Forest algorithm. In order to ensure correct authentication, data features must be consistent. This consistency is provided through the normalization of the PPG signal using maximum-minimum normalization and spline interpolation. The threshold is set by using normalized data and the highest value among the points above the threshold is set as the peak of the period. After establishing candidates of valley with values below the threshold, a shifted period is calculated by designating the last valley next to the following peak to the valley of the period. In order to reduce errors in the one period PPG data, the data is averaged after overlapping. A discrete cosine transform (DCT) is used to extract features from the preprocessed data. Thus, the extracted features are used as the input variables for machine learning techniques such as Decision Tree, KNN, Random Forest. The accuracy of these algorithms is 93%, 98%, and 99% respectively.