利用樸素貝葉斯分類器基於心率變化探測人體行動之方法

Abstract

人體行動探測技術在日益來臨的老齡化社會中透過穿戴設備遠端實時對老年人的行動進行判斷，降低了看護所需的人力成本。傳統的行動探測有利用重力傳感器分析人體肢體運動或透過行動中的心電圖分析心率變異性等方法，這些方法存在著採集資料不方便，同時實驗環境與日常生活存在較大差異的缺點。本論文研究基於長時間採集得到的心率資料計算心率波動作為行動探測特徵的方法。 本研究的實驗方法是利用已普及的心率錶作為資料設備，在日常生活中採集實驗對象的心率資料，按照規則進行行動標註，計算單位時間內的心率的標準差和變化均方根作為描述心率波動範圍和劇烈程度的指標，并以此作為區分各個人體行動的特徵。 樸素貝葉斯模型是一種基於貝葉斯定理和特徵獨立假設的經典的機器學習方法，透過計算一系列特徵在不同分類下的聯合概率分佈選取最大概率實現分類的的目標。原理簡單，構造模型速度快，針對符合特定概率分佈模型的特徵分類效果明顯。同時基於重採樣技術的Bagging法可以實現對不穩定分類的結果的顯著提升，達到模型性能整體優化的效果。 實驗的結果說明本研究所採用的方法和特徵可以有效的辨識人體的行動，針對辨識效果不佳的分類透過重採樣方法有效提升準確率。

The technology on human activity detection can classify the action what the elder do through wearable device in real-time, it is useful to reduce labor costs for caring in an increasingly of aging society coming. There are some thesis like body pattern recognition by G-sensor and heartrate variability analysis by ECG for use. Those method have some disadvantage include inconvenient collection at data or difference between experiment and life. This research is a new method about activity detection based on big data that is collected from life. Experiment used a watch with continuous heartrate sensor to capture participant’s data include heart pulse, steps, calories of all activities whole day. Then calculate the standard deviation and the root mean square of the successive differences between heart pulse in 5 minutes. These features describe heart rate variability on activities and can be the basis of classification. In machine learning, naive Bayes classifier is based on applying Bayes' theorem with strong (naive) independence assumptions between the features and get the result through maximum a posteriori and the conditional distribution of class under features. The thesis is easy to understand , the training can be take in lower time and the classification is effective. Bootstrap aggregating, also called bagging, is a machine learning ensemble meta-algorithm based on resampling designed to improve the stability and accuracy of unstable class. Result shows the method that is described in this research can classify activities effectively using heart pulse variability.