CMARPGA: Classification based on multiple association rules using parallel genetic algorithm pruned decision tree

Abstract

Associative classification is recognized by its high accuracy and strong flexibility in managing unstructured data. However, the performance is still induced by low quality dataset which comprises of noisy and distorted data during data collection. To overcome the influences of low quality dataset, this research proposes a new optimized pruning technique to prune while optimizing the decision tree using genetic algorithm. To achieve the most ideal decision tree, fitness value is weighed using not only the accuracy of the class association rules but also with the size of the decision tree. The size is obtained from the number of nodes of that particular decision tree. However, the fitness value formula is not putting more weightage on the size of the tree as accuracy from a small decision tree will tend to overfit to the training data. Experiments were conducted using databases from UCI machine learning database repository and the results showed that the proposed prediction model is consistently reliable and has good overall accuracy.