<p>Laser-based additive manufacturing (LBAM) has transformed the production of complex metallic components through precise, layer-by-layer deposition. However, porosity defects can compromise the mechanical integrity of printed parts, necessitating effective real-time monitoring and defect detection methods. This study presents a novel, physics-informed framework for in situ porosity classification using shallow learning (SL) models and captured thermal data from a dual-wavelength pyrometer sensor. Unlike deep learning models that require high-resolution large datasets and extensive computational resources, our approach leverages engineered features from multi-orientation (0°, 90°, + 45°, and − 45°) thermal profiles – captured along the laser scan, transverse, and diagonal directions – to characterize melt pool behaviour. We introduce two physics-informed features, melt pool distance (MPD) and aspect ratio of maximum temperature to MPD (ARTM), alongside interpretable statistical feature set. To address the severe class imbalance in defect categories (no-, micro-, and macro- porosity), we apply Synthetic Minority Oversampling (SMOTE) and evaluate model performance using traditional metrics and a novel Classification Deviation Error (CDE) metric proposed to capture minority class misclassification. Our results demonstrate that SL models such as logistic regression achieve high classification accuracy (up to 95%), precision (96%), and recall (95%), and low inference latency, making them suitable for real-time monitoring. The proposed approach offers a computationally efficient and interpretable solution for multi-class porosity detection and paves the pathway for closed-loop quality assurance in LBAM.</p>