Neural Network-Augmented Locally Adaptive Linear Regression Model for Tabular Data

Creating an interpretable model with high predictive performance is crucial in eXplainable AI (XAI) field. We introduce neural network-based regression for tabular data this study. Our proposed uses ordinary least squares (OLS) as a base-learner, and we re-update the parameters of our base-learner by using networks, which meta-learner model. The updates coefficients confidence interval formula. extensively compared to other benchmark approaches on public datasets task. results showed that outperformed models. also applied synthetic measure interpretability, can explain correlation between input output variables approximating local linear function each point. In addition, trained economic discover central bank policy rate inflation over time. As result, it drawn effect rates tends strengthen during recession weaken expansion. performed analysis CO2 emission data, discovered some interesting explanations target variables, such parabolic relationship emissions gross national product (GNP). Finally, these experiments could be applicable many real-world applications where type explainable models are required.