在SVM 的訓練過程中,如何決定目標函數中的參數值與核心函數中的 參數值一直是尚未解決的問題,本論文的目地在於提出一套新的演算法以決定目標函數中理想的C 值與核心函數中理想參數值,不同的C 值與核心函數的參數值會導致有不同的正確率,本論文中核心函數以高斯函數為主。此演算法以模糊理論為基礎,經由模糊規則以推導出理想的C 值與高斯函數中的γ 值,雖然目前已有一些相關論文提出,但其方法與結果尚有相當大的改善空間,本篇論文分析資料分佈情況與核心函數的影響,將其結果歸納於九條模糊規則中以取得其中的C 值與γ 值,並以類神經網路進行學習。經實驗結果證明,本篇論文所提出的方法能有效的產生正確的C 值與γ 值。
In the phase of building SVM model, it is still an unsolved problem of how to decide the optimal parameter values for the cost function and kernel function.Although numerous researches have been proposed to overcome this problem, they were suffered with the problem of much time complexity. Ideal parameter values could increase the accuracy of classification. In this thesis a novel algorithm is proposed to generate ideal parameter values. In this algorithm,overall relations between training patterns are summarized into nine fuzzy rules and fuzzy inference engine is used to generate the ideal parameter values.Besides, fuzzy neural network is used to reach the optimal solution.Experimental results show that our proposed algorithm produces ideal C and γ effectively and outperform other methods.