การประยุกต์ใช้แบ็กพรอพาเกชันนิวรอลเน็ตเวิร์ก เพื่อจำแนกประเภทจากกฎความสัมพันธ์ของประเภท / นภาพรรณ ยิ่งชาญกุล = Applying backpropagation neural networks for classification from class association rules / Napaparn Yingchankul
Associative classification is a promising approach for building accurate classifiers and producing human understandable classification rules. However, the original algorithm, CBA, has some drawbacks during the classifier builder phase. several subsequent algorithms have been invented for better performance. Unlike CBA, this thesis proposes applying backpropagation neural networks to class association rules (CARs) during the classifier construction phase, rather than ranking them. The method applies neural network for classifying in two ways. First is to select a CAR used for classifying new data. The other is to directly use it for predicting the class label of new data. The advantage of our approach is the ability of employing prior knowledge uncovered in the previous CARs discovery phase to help defining the structure of neural networks that would yield more accuracy rate. Compared with C4.5 learning method, the original CBA and CPAR, the experimental results showed that the proposed method gave the bestclassification accuracy
