Prevalence of Fatty Liver among Children under Multiple Machine Learning Models

To analyze the possible factors causing fatty liver in children based on ultrasound data of children in south Texas, and to establish machine learning models of fatty liver in children to provide ideas for the prevention and treatment of fatty liver in children.

The binary classification model of fatty liver problem in obese children in Texas was established under the multiple model. First, we selected important features using the CatBoost algorithm. Second, the best parameters of the algorithm were selected on the training set and the validation set by using the grid search method, and all six models were tested on the test set. The six models then were compared by area under the curve value, precision, accuracy, recall rate, and F1 score in a model evaluation. Then, two algorithms, logic regression and CatBoost, were selected to establish prediction models of fatty liver disease in children.

We selected body mass index, height, liver size, kidney volume, glomerular filtration rate, and liver diameter as the features used in the machine learning model. The prediction models we chose showed that children with higher body mass index at the same age tended to have a greater probability of fatty liver.

Based on the analysis of the results of the two prediction models established by logistic regression and CatBoost, we determined that the mean probability of fatty liver in severely obese children was between 74.47% and 92.22%, 73.45% and 85.41% in obese children, and slightly higher in boys than in girls, with a mean difference of 3.00% to 3.95%.