A Hybrid ANFIS-Gradient Boosting Frameworks for Predicting Advanced Mathematics Student Performance

This paper presents a new hybrid prediction framework for evaluating student performance in advanced mathematics, thus overcoming the inherent constraints of classic Adaptive Neuro-Fuzzy Inference Systems (ANFIS). To improve predictive accuracy and model interpretability, our method combines ANFIS with advanced gradient boosting techniques, namely XGBoost and LightGBM. The proposed framework integrates fuzzy logic for input space partitioning with localized gradient boosting models as rule outcomes, effectively merging the interpretability of fuzzy systems with the strong non-linear modeling capabilities of machine learning. Comprehensive assessment reveals that both the ANFIS-XGBoost and ANFIS-LightGBM models substantially exceed the traditional ANFIS in various performance parameters. Feature selection, informed by SHAP analysis and XGBoost feature importance metrics, pinpointed essential predictors including the quality of previous mathematics education and core course grades. Enhancements in regression measures further highlight the effectiveness of the hybrid methodology. The findings indicate that the suggested framework provides a reliable and effective alternative for educational institutions to forecast student success, guide focused interventions, and enhance learning outcomes in advanced mathematical fields.

کلیدواژه ها:

Adaptive Neuro-Fuzzy Inference Systems ، Feature selection ، Hybrid ANFIS-XGBoost model ، Hybrid ANFIS-LightGBM model