A novel Q-learning-based FKG-Pairs approach for extreme cases in decision making

The decision-making problems based on fuzzy inference systems have received much attention from the worldwide scientific community. The M-CFIS-FKG model is considered one of the best models to solve classification problems based on uncertain and amplitude input datasets. It can infer and find the output labels of new samples that are not in the fuzzy rule base. Recently, the FKG-Pairs model has been considered an extension of FKG in the M-CFIS-FKG model by combining attribute pairs to infer and find the output labels in the cases of input datasets with incomplete gathering. It has overcome the limitation of the M-CFIS-FKG model. However, with real-time large input datasets or too-small training datasets, the FKG-Pairs model has also revealed limitations as it takes too much time to compute, and the accuracy is still relatively low. This paper has proposed a decision-making model in extreme cases (called FKG-Extreme) by using the Q-learning-based FKG-Pairs approach to enrich the fuzzy rule base after each time step with the cumulative mechanism of new rules. The proposed FKG-Extreme model has overcome the FKG-Pairs model's limitation in the extreme case. It has significantly improved the system's accuracy, while the computation time is acceptable. To validate the proposed model, we conducted experiments based on the standard UCI datasets, and the experimental results demonstrated that the system's performance in terms of accuracy is superior to the other reliable models in case of too small training data. Furthermore, the results of the two-way ANOVA also proved that the FKG-Extreme model is better than the FIS and FKG-Pairs models.