Student Performance Evaluation Using an Interpretable Genetic–Fuzzy Framework with Rule-Base Reduction
Uncertainty in instructors' judgments and qualitative evaluations is difficult to capture in traditional student performance evaluations centered on descriptive statistics and averages. While traditional fuzzy systems are capable of handling this kind of ambiguity, they become less efficient and harder to understand as the number of evaluation variables rises since their rule bases are exponentially larger. Exam results may be used to evaluate students' performance, and this research suggests a genetic-fuzzy architecture with rule-base reduction that can be understood. The model was trained in MATLAB using the following parameters: 50 nodes, 25 generations, 0.8 crossover probability, 0.2 mutation rate, and roulette-wheel selection; Mamdani inference; centroid defuzzification; and a genetic algorithm. The non-optimized fuzzy method's MAE is 0.0208 and the genetic-fuzzy method's MAE is 0.0101, according to recalculation utilizing the 25 records shown in Table 1. These values are unique to the descriptive versus creative 25-record comparison and should not be used interchangeably with the separate function membership creative sensitivity results from Table 4. The original basic set of 125 rules was pared down to 46 rules. The leakage-free train-validation-test methodology is defined in this version, and final generalization estimations must be given from it.