International Journal of Clinical Case Reports, 2025, Vol.15, No.5, 209-218 http://medscipublisher.com/index.php/ijccr 214 In terms of computing speed, improved artificial neural networks and hybrid models generally run faster than traditional deep learning or random forests, and are more suitable for real-time processing or large-scale tasks (Khalid et al., 2023). However, deep learning models generally require stronger computing resources and more training data. Although the simple model runs quite fast, its prediction effect may not be very good when encountering more complex problems (Akter et al., 2021; Chittora et al., 2022). 5.3 Case study: application effects of different models in chronic disease prediction Some research cases of chronic diseases have demonstrated the practical performance of different methods. For example, the optimized artificial neural network with feature selection performs better than other models in the prediction of breast cancer, diabetes, heart disease, hepatitis and kidney disease, with an accuracy rate of up to 99.67%, and has less running time than random forest, SVM or deep learning models (Figure 2) (Rashid et al., 2022). In the prediction of chronic kidney disease, the hybrid model combining decision tree, gradient boosting and random forest achieved extremely high or even near-perfect accuracy, demonstrating the value of the ensemble algorithm (Chittora et al., 2022; Khalid et al., 2023). Figure 2 Data processing time for cancer, diabetes, heart, hepatitis and kidney disease (Adopted from Rashid et al., 2022) In some cases, if only a few clinical predictors are used, the performance of logistic regression is similar to that of more complex machine learning models. This indicates that when choosing a model, the characteristics of the data and clinical needs should be combined. Deep learning models, such as LSTM and MLP, perform strongly in the early prediction and risk identification of chronic kidney disease, especially when large-scale and multimodal data are used (Akter et al., 2021; Sawhney et al., 2023). These findings remind us that models should be selected to be the most appropriate based on the type of disease, data conditions and actual application scenarios. 6 Discussion and Challenges 6.1 Applicability of different models in actual chronic disease prediction scenarios The application of chronic disease prediction models in clinical practice depends on specific scenarios, data conditions and the complexity of the disease. Traditional statistical models, such as logistic regression and Cox regression, remain useful in cases where the data volume is small and structured due to their ease of understanding and integration into clinical processes (Battineni et al., 2020). However, as medical data becomes increasingly complex and high-dimensional, machine learning and deep learning models, such as random forests, XGBoost, and neural networks, perform better in the early prediction and risk stratification of diabetes, cardiovascular diseases, and chronic kidney diseases (Delpino et al., 2022; Rashid et al., 2022). In addition, multi-task and multi-modal learning methods can simultaneously predict multiple diseases and utilize different data sources, and are receiving increasing attention in actual medical scenarios (Kim et al., 2023).
RkJQdWJsaXNoZXIy MjQ4ODYzNA==