This study presents a comparative analysis between rule-based (Forward Chaining) and machine learning (Decision Tree, Random Forest) approaches for classifying mental health care needs. Utilizing a Kaggle dataset (>290,000 rows), the methodology enforces a strict pipeline separation. Forward Chaining rules were extracted via statistical rule-mining from training data without SMOTE, whereas machine learning models incorporated SMOTE. Models were evaluated using accuracy, precision, recall, F1-score, and confusion matrix metrics. Key contributions include a sound methodological pipeline separation, rule derivation from empirical data, and an analysis of decision transparency alongside SMOTE's impact on mild class imbalance. Results demonstrate that machine learning models (with SMOTE) yield higher and more consistent performance across all metrics compared to Forward Chaining. These findings serve as an indicative decision-support tool for early mental health screening rather than a clinical diagnosis substitute.