應用類神經網路於統計機率分配辨識之研究

Abstract

在統計資料分析中，必須先已知樣本資料的統計機率分配為何，才能瞭解樣本資料所提供資訊，以利於作出進一步的正確決策分析。傳統上的統計機率分配辨識方法以非參數統計中的適合度檢定為主。然而，適合度檢定亦有其限制，如：樣本個數太少無法作出精確的辨識、分組組數影響辨識結果。有鑑於此，本研究提出應用類神經網路來建構一套統計機率分配之辨識方法，並與傳統統計適合度檢定作比較。此外，本研究舉一數據分析，來說明所提出之應用類神經網路的辨識方法，結果顯示本研究所提出的方法具有高度辨識正確率及時效性。

In Statistic data analysis, we first have to know what kind of statistic probability distribution the data obeys; therefore, we can understand what information the data provides and make a right decision making. In general, a non-parameteric test for goodness of fit is used in distribution recognition. However, there are several constraints and limitation. For example, due to the inadequacy of sample, precise recognition can not be made. In addition, the numbers to divide into groups influence the results. This research presents an effective procedure capable of recognition statistic probability distribution by employing neural network. The proposed approach is compared with non-parameteric test for goodness of fit. A data analysis involving the distribution recognition demonstrates high recognition rate and effectiveness by our proposed approach.