穿戴式裝置資料在羽球訓練上可行性分析

本研究目的為穿戴式感測器裝置接收的資料是否能夠偵測羽球球員的等級與揮拍動作種類。我們透過一組實際實驗資料的分析，一方面介紹感測器資料與示範分析步驟，另一方面也討論分析過程中所面臨的挑戰，可提供其他統計研究人員作為應用與參考。我們觀察到本組感測器資料不穩定，統計分析的前處理為重要且關鍵的程序。本研究的前處理包括處理NA值、判斷與刪除離群值、資料正規化以及偵測各揮拍動作。經過前處理後，各揮拍動作的序列為高維資料，我們考慮運用兩種維度縮減的方式來降維以簡化後續分析的複雜度與計算。經過降維後的資料以單變量變異數分析做差異性分析。有鑑於實驗在各因子組合下的不平衡特性以及不符合常態分配假設等因素，我們也搭配拔靴法來判斷球員等級、球員、球種與之間的交互作用是否達統計顯著。經過本研究初步分析發現該組感測器資料的確能顯著檢測球員、球員等級以及揮拍動作之間的差異，故未來穿戴式裝置的確可運用在羽球推廣、教學或訓練上，例如適合普羅大眾的商業化健身遊戲裝置、羽球課程的設計與學員評量或研究羽球的運動科學研究等。

關鍵字

穿戴式裝置；資料前處理；維度縮減；變異數分析；拔靴法

並列摘要

This research is to study the feasibility of a wearable device in differentiating the level of a badminton player and the movement of stroke type that the player hits. On one hand, we introduce the data collected from a sensor of a wearable device and demonstrate the analytical procedure via a real example data. On the other hand, we discuss the difficulty and challenge in the analysis, which may motivate more advanced researches. The exampled sensor data is noisy, the preprocessing is important and crucial. In this study, the preprocessing includes handling the NA values, determining and deleting the outliers, normalizing the data and detecting hitting. After preprocessing, we obtain sequences of the movement that a player hits the shuttlecock. The sequential data is highly dimensional, two types of dimension reduction are applied to reduce the complexity of follow-up statistical analysis. The ANOVA of a nested factorial design is employed on the reduced data to test the significance of player’s level and movement type. In the light of an unbalanced experiment and obvious failure of normality assumption, the bootstrapping method is considered in ANOVA. Through this initial analysis we find that sensor’s data can significantly detect the players’ level and movement type. Therefore, in the future there is a great potential to develop a wearable device on popularization, teaching or training of badminton sport.