Predicting human body composition using a modified adaptive genetic algorithm with a novel selection operator

Changes to human body composition reflect changes in health status to some extent. It has been recognized that these changes occur earlier than diseases. This means that a reasonable prediction of body composition helps to improve model users' health. To overcome the low accuracy and poor adaptability of existing human body composition prediction models and obtain higher efficiency, we proposed a novel method for predicting human body composition which uses a modified adaptive genetic algorithm (MAGA).

Firstly, since there are many parameters for a human body composition model, and these parameters are associated, we designed a new parameter selection approach by combining the improved RReliefF method with the mRMR method. Following this, selected parameters were used to establish a model that fits body composition. Secondly, in order to accurately calculate the weight of parameters in this model, we proposed a solution which used a modified adaptive genetic algorithm, taking advantage of both roulette and optimum reservation strategies. This solution also has an improved selection operator. Thirdly, taking the percentage of body fat (PBF) as an example of body composition, we conducted experiments to compare performance between our algorithm and other algorithms.

Through our simulations, we demonstrated that the adaptability of the proposed model is 0.9921, the mean relative error is 0.05%, the mean square error is 1.3 and the correlation coefficient is 0.982. When compared with the indexes of other models, our model has the highest adaptability and the smallest error. Additionally, the suggested model, which has a training time of 28.58s and a running time of 2.84s, is faster than some models.

The PBF prediction model established by MAGA has high accuracy, stronger generalization ability and higher efficiency, which could provide a new method for human composition prediction.