Enhanced 1D Convolutional Neural Network for superior obesity risk assessment

Obesity has emerged as a significant public health challenge worldwide, necessitating advanced predictive models for early intervention. It has now become extremely important to predict the obesity risk level. In this work, we employ deep learning approach, One Dimensional Convolutional Neural Network (1DCNN), on extensive datasets containing demographic information, dietary patterns, and physical activity levels. The purpose of the investigation is to predict the risk factors associated with different levels of obesity. The different layers of the proposed 1DCNN model are presented with their significance enabling it to learn feature interactions hierarchically, and sliding the same kernel over the input drastically reduces the number of trainable parameters, which differentiates the proposed model from other traditional models. Our model asses the obesity prediction using the dataset of 2111 instances with 17 attributes available in the public domain and compares it with nine traditional machine learning models. The proposed model has shown 98 % training accuracy, 97 % validation accuracy and 99 % testing accuracy. The model identifies the most influential risk factors for obesity, and utilizes predictive analytics to estimate the probability of occurrence. The saliency maps, and SHAP depiction show that weight, family, FCVC, height, and age are more influential risk factors for predicting obesity and MTRANS and CALC are not influential factors.