財務危機預測問題(Financial distressed prediction problem)已經經過了長時間且廣泛的討論，而本研究主旨在美國上市公司資料集中擴展FDPP的研究方向，過往的學者大多運用財務特徵來進行FDP，而本實驗希望能找出除了財務特徵之外新的特徵能夠提升預測財務危機的表現，且因特徵資料型態的不同也會影響預測的結果，過去已經有學者運用應計項目(Accruals)來進行FDP，但使用的應計項目並不全面，或是著重的問題不再FDP而是在盈餘管理(Earning management)，且所使用資料型態都為年資料，因此本實驗著重於運用所有的應計項目(Accruals)，跟使用時間序列(Time series)季資料來進行研究，之後會輔以特徵降維來降低維度以提高特徵表現跟進行特徵權重分析。;The financial distressed prediction problem(FDPP) has been discussed for a long time and extensively. The main purpose of this thesis is to focus on US listed companies data to extend FDPP research direction. Most of previous scholars and researcher used financial ratio(FR) to do the prediction. This thesis is hopes to find out new feature besides financial ratio which can improve the performance of FDP result. And we know difference data type will also affect the prediction result. In the past, some scholars had used accruals as feature to do prediction, but its accruals are not comprehensive, or the research question is not focus on FPD but Earning management, and also the data type are year data. Therefore, this thesis focuses on use comprehensive accruals and using time series quarter data to do the research. After all we will dimension reduction to reduce dimensions to improve feature performance and perform feature weight analysis.
