Disability Prediction in the Elderly by a Convolutional Neural Network Model With Convolutional Block Attention Module

With the increasingly rapid aging of the global population, the population of middle-aged and older people with disabilities has been increasing, and thus disability has gradually been highlighted as an important issue for the middle-aged and elderly population. In this article, a 1-D convolutional neural network (1DCNN) model incorporating the convolutional block attention module (CBAM), i.e., the 1DCBAMCNN model, is designed and applied to the risk assessment of disability in the elderly. The 1DCNN effectively captures both local and global features within the data through convolutional operations. By directly analyzing tabular data related to disabilities in the elderly, it circumvents the information loss and increased computational complexity associated with dimensionality transformation. Additionally, the CBAM, which comprises channel attention and spatial attention modules, facilitates fine-grained feature selection and weighting of the input data. The integration of these two components not only leverages the robust feature extraction capabilities of the 1DCNN but also utilizes the attention mechanism of CBAM to delve deeper into the abstract characteristics of the data, significantly enhancing the model's sensitivity to data features and effectively improving predictive performance. The model proposed in this article (1DCBAMCNN) experiments using the China Health and Retirement Longitudinal Study (CHARLS) dataset, and the model is compared with deep learning methods such as convolutional neural network (CNN), artificial neural network, and long short-term memory, as well as with the CNN model that incorporates the squeeze-and-excitation module, the efficient channel attention module, and the self-attention, respectively. The experimental results show that 1DCBAMCNN has better prediction results on the test set compared to other models, with accuracy and F1 score reaching 92.46% and 80.25%, respectively. The 1DCBAMCNN model presented in this article can more effectively predict disabilities within the elderly population, providing a basis for diagnosis and treatment.