Prediction of body fat percentage using a new hybrid intelligent feature selection: ANN-KGA & ANN-IKGA

Before initiating obesity treatment, it is essential to determine body fat percentage (BFP), a measure that cannot be assessed through conventional weighing alone. To address this, specialized “Body Analyzers” have been developed; however, their high cost encourages the search for more practical and cost-effective alternatives. In this study, 16 machine learning algorithms (six linear and 10 non-linear) were evaluated for BFP prediction and compared with two hybrid evolutionary frameworks combining an Artificial Neural Network (ANN) with a K-means Genetic Algorithm (ANN-KGA) and an Improved K-means Genetic Algorithm (ANN-IKGA). To ensure methodological rigor and prevent optimistic bias, a strict nested cross-validation strategy (five outer folds × three inner folds) was employed during feature selection and model evaluation. To account for the stochastic nature of the genetic algorithm, the entire nested cross-validation procedure was repeated 15 times with different random seeds. The ANN-KGA model achieved a mean outer-fold R² of 0.7522 (95% CI [0.6963–0.8081]) with a mean Root Mean Squared Error (RMSE) of 3.9651 (95% CI [3.5453–4.3849]) across the 15 runs. The proposed ANN-IKGA model demonstrated strong predictive performance, achieving a mean outer-fold R² of 0.7858 (95% CI [0.7495–0.8222]) and a mean RMSE of 3.7182 (95% CI [3.1777–4.2586]) across the 15 independent runs. Statistical comparison against baseline models using paired t-tests indicated significant improvement ( p < 0.05) for both evolutionary approaches. These findings demonstrate that the proposed ANN-IKGA framework enhances predictive accuracy and generalization performance under strict nested validation. The results highlight the effectiveness of evolutionary feature optimization in improving BFP estimation using anthropometric measurements, providing a reliable and cost-effective alternative to specialized body analysis equipment.