在目前醫療發達的社會中，人類壽命愈來愈長，而臺灣目前正邁向老年化社會，同時也面臨少子化的問題。根據統計顯示，目前每6.6 個工作人口要扶養1位老人，而行政院主計總處預估到2031 年時，每2.2 個工作人口就要扶養1 位老人，因此，如何有效利用資訊科技協助照護老人已成為目前一個重要的議題。在老年照護中，最容易且最怕發生的就是「跌倒」，因為跌倒後可能產生意識不清等情形，因此，若能利用資訊科技正確地進行跌倒偵測，即可有效率地在第一時間發出求救訊息。
過去在老人跌倒偵測的議題中，必須在老人身上放置許多相關的偵測設備，但這些偵測設備多數過於昂貴或笨重，因此，本研究改以最容易取得的「手機」作為偵測設備。目前臺灣人手一機，若使用手機就不需另購其他昂貴設備，同時再搭配手機內的三軸重力加速度感測器（G-Sensor）所感測出的X、Y、Z 三軸的數值變化，以判斷使用者的行動狀態。本研究將透過智慧型手機的G-Sensor 所感測出的X、Y、Z 三軸的變化搭配資料探勘的技術，作為偵測老人跌倒的判斷依據，結果顯示，本研究所提出的方法在老人跌倒偵測的正確率高達96% 以上。

According to the inference of Taiwan Ministry of the Interior, 2.2 working people will have to take care of one elderly in 2031. It is, therefore, important to facilitate a caring environment to safe guard the wellbeing of the aged. However, according to the previous studies, the most important issue in taking care of the elderly is to avoid physical falls which may affect the overall health of the elderly. To prevent the unexpected falls, many researchers used technology products to construct the fall detection system. However, most of the technology products are either expensive or massive in size. To solve the cost and size issues, this study constructed and implemented a cloud elderly fall detection system based on a wearable device and a classification model. The proposed system, firstly, used the G-Sensor of a smartphone to detect the activity patterns of the elderly. Subsequently, the proposed system would use the classification models of data mining technique to classify and to predict the activity patterns of the elderly. Since the data collected from G-Sensor was a set of time series data, the proposed system used the sliding window model to perform data pre-processing to enhance the accuracy of the classification. According to the results, the classification accuracy rate of the proposed detection system for elderly fall achieved as high as 96%.