Framework for Optimizing The Performance of Fuzzy Grid Partition for Rules Generation

In the development of fuzzy machine learning, fuzzy systems that were originally designed manually by experts, in some cases, have switched to models that are built automatically. For a fuzzy system to function properly, its fuzzy rules must be determined precisely. Fuzzy rules obtained from the knowledge of experts are better. Still, it is difficult to model the logical implications of humans because of the unavailability of a modelling process framework and expensive expert fees. Fuzzy rules can be generated using the grid-partition approach, where the grid-partition approach can overcome the problem of a too fine grid. However, the number of generated rules increases significantly in each iteration, so another approach is needed to reduce the number of rules. This research is to describe a framework to optimize the performance of the fuzzy grid partition method in generating rules for classifying datasets. The generated fuzzy rules apply the hybrid fuzzy grid partition method and the Rough Set method called the Rough Set Fuzzy Grid Partition (RS-FGP) method. The Rough Set method is used at the beginning of the rule formation, and the fuzzy grid partition method generates the rules. Testing the success of the RS-FGP method by looking at the level of accuracy of the fuzzy rules generated in classifying data sets. The optimal number of fuzzy rules was obtained from the experiments with an accuracy rate of 83.33%. Accuracy results show an increase of 16.66% from the classic Fuzzy Grid Partition (FGP) method.