Article: A frequency-domain enhanced fusion framework for predicting students' physical health risks Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.18 pp.41 - 52 Abstract: Traditional health evaluation methods mainly use single-viewpoint data representation. Such methods are hard to learn the complex interrelationships between physiological and behavioural factors. This article introduces a dual-branch model that integrates spatial features and frequency features. The original features are processed by multi-layer perceptrons, while the frequency features are learned through the fast Fourier transform and modelled using one-dimensional convolutional neural networks. Finally, the outputs of the two branches are fused to generate the final prediction result. This study conducted experiments on a student health dataset that includes physiological, behavioural and psychological attributes. The proposed model was compared with several classic machine learning methods. The results showed that the proposed method achieved the best overall performance, with an accuracy rate of 0.900 and an F1 score of 0.8977. Further analysis using methods such as confusion matrix, ROC curve and feature visualisation confirmed its excellent performance. Inderscience Publishers - linking academia, business and industry through research

Title: A frequency-domain enhanced fusion framework for predicting students' physical health risks

Authors: Weitao Guo; Baoqin Yan

Addresses: Anyang Preschool Education College, West of the intersection between Zhonghua Road (South Section) and Renhe Road in Tangyin County, Anyang City, Henan Province, 455000, China ' Fuzhou University Zhicheng College, No. 50, Yangqiao West Road, Gulou District, Fuzhou City, Fujian Province, 150071, China

Abstract: Traditional health evaluation methods mainly use single-viewpoint data representation. Such methods are hard to learn the complex interrelationships between physiological and behavioural factors. This article introduces a dual-branch model that integrates spatial features and frequency features. The original features are processed by multi-layer perceptrons, while the frequency features are learned through the fast Fourier transform and modelled using one-dimensional convolutional neural networks. Finally, the outputs of the two branches are fused to generate the final prediction result. This study conducted experiments on a student health dataset that includes physiological, behavioural and psychological attributes. The proposed model was compared with several classic machine learning methods. The results showed that the proposed method achieved the best overall performance, with an accuracy rate of 0.900 and an F1 score of 0.8977. Further analysis using methods such as confusion matrix, ROC curve and feature visualisation confirmed its excellent performance.

Keywords: multimodal learning; health risk prediction; college students; frequency features.

DOI: 10.1504/IJRIS.2026.155468

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.18, pp.41 - 52

Received: 23 Apr 2026
Accepted: 27 May 2026
Published online: 03 Aug 2026 *

Title: A frequency-domain enhanced fusion framework for predicting students' physical health risks

Keep up-to-date