Thermal imaging method to evaluate childhood obesity based on machine learning techniques

Abstract The purposes of the study were (i) to determine the potential of thermal imaging to assess the difference in the thermal pattern in various body regions of studied population; (ii) to compare the performance of feature extraction, feature fusion, feature ranking and feature dimension reduction (PCA) in classification of obese and normal children using different Machine learning algorithms. About 600 thermograms were obtained from various regions such as abdomen, finger bed, forearm, neck, shank and gluteal region for the studied population. Fifteen statistical textual features were extracted from the six regional thermograms followed by implementing feature fusion with SIFT and SURF algorithm. The PCA method provides the best classification accuracy for SVM (98%) followed by Naïve Bayes and Random Forest (97%). Thus, the regional thermography and computer aided diagnostic tool with machine learning classifier could be used as a basic non‐invasive prognostic tool for the evaluation of obesity in children.