學習向量量化網路之模擬研究

Abstract

本研究採電腦模擬探討不同訓練範例樣本數、學習速率及隱藏層個數對學習向量量化網路分類正確率之影響。在實驗設計上，先產生三種混合樣本資料，以比較在各種不同的實驗設計下，學習向量量化網路分類正確性之差異。結果發現不論為何種混合樣本資料，分類正確率會隨著訓練範例樣本數的增加而呈現遞增的趨勢。在學習速率方面則有一共同點為當學習速率0.1時，其模式之分類效果都是最差的；其餘學習速率的分類表現並無規則可循，亦無表現突出者。但當以不同隱藏層處理單元數進行模擬試驗所得結果，以隱藏層個數為2時，分類正確率都是最佳的。 The purpose of this study was to explore the effect of the different training samples, learning rate and numbers of hidden layers on the classified accuracy when using Learning vector Quantization to analysis. When the number of the training samples were on the two numbers, 200 and 250, the classified accuracy was irregular, and the difference between these two number was not significant. As a result, the classified accuracy of training samples were increasing while the training samples were increasing. By using the simulation testing method to compare the accuracy of learning rate, in any kind of mixed samples, when the learning rate was on 0.1, the classified accuracy was the worst; however the classified accuracy from other learning rates was irregular. When the number of hidden layers was 2, the classified accuracy was the best in any kind of mixed samples.