改良型灰色狼最適化アルゴリズム-ランダムフォレストモデルに基づく小児健康予測の研究

Children's health is a crucial indicator for evaluating public health standards and social development. With the changes in lifestyles and the increasing complexity of environmental factors, health issues such as childhood obesity, allergic diseases, and respiratory infections have become increasingly prominent. Traditional health assessment methods that rely on periodic physical examinations and questionnaires have limitations such as data lag and subjectivity. This paper proposes a hybrid model (IGWO-RF) that integrates the Improved Grey Wolf Optimization (IGWO) algorithm with Random Forest (RF) to enhance parameter optimization and interpretability in pediatric health prediction. Firstly, a RF prediction model is constructed based on children's physical examination data. Secondly, dynamic weight strategy and elite retention mechanisms are introduced to improve the Grey Wolf algorithm for optimizing RF hyperparameters. The model uses Shapley Additive Explanations (SHAP) values to interpret key factors affecting health. Experimental results show that the IGWO-RF model achieves an accuracy of 92.1% and a F1-score of 90.8%, which significantly outperforms the traditional RF model (85.3%) and the particle swarm optimization-RF model (88.7%). SHAP value analysis identifies body mass index (contribution value 0.32), daily exercise time (0.21), and particulate matter 2.5 exposures (0.18) as the top 3 determinants of children's health. It should be noted that these associations reflect the model's feature importance based on SHAP values and do not imply causal inference. The IGWO-RF model is potentially useful for children's health risk stratification, pending external validation, with excellent predictive performance, and its methodological framework can be extended to early warning systems for other chronic diseases after further testing.