本論文之目的乃是期望以類神經網路之倒傳遞網路與統計分析方法(區別分析、LOGIT分析模式)架構分類模型，並就實證結果進行比較評估它們之間分類效果的優劣、適用性與限制。
於範例模擬測試結果顯示，在訓練樣本建構模式上，以類神經網路模式整體來說總正確率最高，模式配置最適當， LOGIT分析次之，區別分析表現最差;在測試樣本則以LOGIT分析最佳，類神經網路次之，區別分析不佳。在建構模式所花費時間上以類神經網路模式最多，LOGIT分析模式次之，區別分析模式最少。並且類神經網路發生過度適當(Overfitting)的風險則較另兩種方法高。 The aims of this thesis expect to construct classification models by using Back Propagation Network of Neural Network and Statistical Analysis Method (LOGIT, Discriminant Analysis). Besides, We compare the advantages, disadvantages, restrictions, adaptations among these methods.
The results come out by simulating many instances. The construction of training samples witness that hit ratio of Neural Network is the highest and the fitted, LOGIT is second and Discriminant Analysis is the worst. The construction of testing samples witness that hit ratio of LOGIT is the highest and Neural Network is second, Discriminant Analysis is the worst. Neural Network takes the longest period to construct these models, LOGIT is second and DiscriminantAnalysis is third. In addition Neural Network might occur more rate of overfitting than the others.
Relation:
第六屆國際資訊管理學術研討會論文集Proceedings of the 6th International Conference on Information Management，頁125-132
