Diversification of fuzzy association rules to improve Prediction Accuracy

Reliability of the Fuzzy Association Rules (FARs) extraction is a challenging research in knowledge discovery and data mining. Reliability refers to the trade-off between the prediction accuracy and the rules diversity. In this paper, an approach called Diverse Fuzzy Rule Base (DFRB) is proposed to extract the FARs which are used later to predict the future values. This approach also aims to ensure high quality and diversity of the FARs. This is achieved through four phases: firstly, the integration of Fuzzy C-Means (FCM) and Multiple Support Apriori (MSapriori) algorithms are applied to extract the FARs. The second phase calculates the correlation values for these FARs, and performs an efficient orientation for filtering FARs as a post-processing method. In the third phase, the FARs diversity is maintained through the clustering of FARs, based on the concept of the sharing function technique used in multi-objectives optimization. After that, the best and the most diverse FARs are evaluated and then stored in the Knowledge Base (KB). Finally, these FARs, stored in the KB, are utilized within the Fuzzy Inference System (FIS) for prediction. Experimental results for two case studies have shown that the proposed DFRB approach predicted the future values effectively, thus, outperforming the existing work.