運用C4.5決策樹分析失眠症狀

Applying C4.5 decision tree to analyze insomnia symptoms

目標：失眠是常見的主觀症狀，會影響生活品質及身體健康，本研究目的為運用C4.5決策樹工具，以找出影響失眠問題關鍵重要因素，以供相關人員參考。方法：本研究收集2015年1月至12月間到台中市某區域醫院接受全身健康檢查之20歲以上之民眾為樣本個案，共收集有效樣本1,223人，其中，男性佔756人（61.8%）。研究使用結構性問卷，有人口統計基本資料、失眠症狀、生活型態、憂鬱及泛焦慮症狀等資訊。本研究以SPSS20.0分析資料，憂鬱、焦慮量表及總量表之內在一致性（Cronbach’sα）值分別為0.78、0.78及0.86。本研究進一步使用Weka3.8.1之「C4.5決策樹」以產生規則以了解哪些因素的組合可能容易導致失眠症狀。結果：入睡困難、夜眠中斷、早醒、醒後疲倦、嚴重失眠問題等各種睡眠障礙因素之間的彼此相關性高；男性與入睡困難、夜眠中斷問題的相關性比女性高；睡前喝水、睡前喝酒、飲咖啡習慣、飲茶習慣皆與嚴重失眠問題正相關；安眠藥與各種睡眠障礙正相關；民眾擁有職業可能會改善入睡困難與夜眠中斷的問題；家庭壓力、工作壓力與其他壓力皆可能會產生睡眠障礙；運動習性的養成，不管是增加運動頻率或是增加運動量皆有助於部份民眾降低失眠障礙；睡前喝酒在短期內有助於克服睡眠問題；戒除飲咖啡習慣對部份民眾將有助於減緩失眠障礙；服用藥物對身體除了可能會產生各種副作用外，亦可能會對部分民眾造成失眠問題；抽煙習慣亦會對部分民眾產生睡眠障礙，擁有職業可降低部份民眾的失眠問題。結論：本研究使用決策樹針對不同類型的民眾進行統計分析以產生規則，發現養成運動習性、少抽煙、少飲咖啡、少服藥物、從事職業活動、睡前避免喝水等，有助於部份民眾改善失眠問題。

Objectives: Insomnia is a common and subjective symptom that influences individuals' quality of life and health. The goal of this research is to apply C4.5 decision tree to identify the key factors for the reference of relevant personnel. Methods: Using a structured questionnaire, we collected physical examination data of 1223 individuals (756 [61.8%] males) older than 20 years attending a Taichung regional hospital. The questions were related to basic demographic information, insomnia symptoms, life style, and mood (depression, anxiety, etc.). This research used SPSS 20.0 to analyze the data. Cronbach's α for depression scores, anxiety scores, and total scores was 0.78, 0.78, and 0.86, respectively. The C4.5 decision tree in Weka 3.8.1 were applied to generate a rule of which combination of factors are most likely to generate insomnia symptoms. Results: “Difficulty falling asleep,” “Interrupted night sleep, “ “Wake up early,” “Wake up tired,” and “Serious insomnia problem” were highly correlated. Males were more likely to score high on “Difficulty falling asleep” and “Interrupted night sleep.” “Drinking water before sleeping,” “Drinking alcohol before sleeping,” “Habit of drinking coffee,” and “Habit of drinking tea” were positively correlated with “Serious insomnia problem.” “Sleeping pills” was positively correlated with each kind of insomnia symptom. “Difficulty falling asleep” and “Interrupted night sleep” were less likely in individuals who were employed. Family pressure, work pressure, and other pressure, may also generate insomnia symptoms. Exercise was associated with reduced insomnia symptoms, irrespective of the exercise frequency and exercise volume. “Drinking alcohol before sleeping” could overcome insomnia problems over the short-term. “Abstaining from drinking coffee” allowed some individuals to reduce insomnia symptoms. Using medicine not only generated various side effects, but may also have generated different insomnia symptoms in some individuals. Smoking can also lead to insomnia symptoms in some individuals, while employment can decrease insomnia symptoms. Conclusions: This study used a decision tree to generate rules on insomnia based on data obtained from a cohort of individuals. We found that taking part in exercise, stopping smoking, drinking coffee, taking medicine, and drinking before sleeping, as well as having employment are likely to improve insomnia symptoms in many individuals.