Lightweight Neuro-Symbolic Framework for Tabular Data Classification Using Rule-Guided Machine Learning

This paper presents a lightweight neuro-symbolic framework designed to improve both prediction performance and interpretability in tabular data classification tasks. The proposed approach combines machine learning with rule-based reasoning to enhance decision consistency. A Random Forest model is used for predictive learning, followed by rule extraction using a decision tree. These rules are then applied in a refinement stage to correct inconsistent predictions. Experimental results demonstrate that the proposed method achieves improved performance while maintaining interpretability, making it suitable for real-world applications in domains such as education, healthcare, and finance.