Neural Network-Based Diagnosis of Abnormal Waist-to-Hip Ratio Values

The combination of anthropometric measurements, such as skinfolds and body circumferences, with artificial intelligence techniques like neural networks has been utilized to indirectly diagnose abnormal levels of anthropometric indices. Artificial neural networks (ANNs) have proven to be valuable in identifying abnormal levels of indices like body mass index and body fat percentage, which are indicators of the risk of obesity-related complications. By training ANNs on diverse datasets, these models can learn patterns and relationships between indices and various factors. The objective of this research is to explore the potential of neural networks in diagnosing abnormal levels of Waist-to-Hip ratio (WHR) using anthropometric variables. To achieve this, a comprehensive database consisting of 1978 individuals with 28 distinct measurements will be analyzed utilizing ANNs and advanced techniques like Monte Carlo cross-validation. This rigorous evaluation and validation process ensures the robustness and reliability of the ANNs models, thereby enhancing the accuracy and effectiveness of diagnosing WHR abnormalities. The findings suggest that a smaller training set may lead to slightly less accurate classification, but the model still performs well with an F1 score above 0.76. The classifier demonstrates high sensitivity (78.8%) in detecting individuals with impaired WHR and specificity (84.1%) in identifying those without impaired WHR. The negative predictive value (86.7%) highlights the classifier's reliability in ruling out individuals without impaired WHR, while the positive predictive value (75.3%) indicates its effectiveness in identifying those with impaired WHR. The low standard deviations across all metrics emphasize the classifier's consistency and robustness.