Hybrid CNN-SVM model for enhanced early detection of Chronic kidney disease

Chronic kidney disease (CKD) is a common disease with a serious prognosis that usually progresses without symptoms until the later stage resulting in extensive organ damage and end-stage kidney disease. Effective intervention at the right time could be life-saving. However, it was noted in the literature that existing machine learning methods had problems such as overfitting and slow computational speed, as well as an inability to deal with class imbalance. Therefore, this research proposed a hybrid convolutional neural network (CNN) − support vector machine (SVM) model to improve the CKD prediction accuracy. The combination of CNN for feature extraction with SVM for classification improved performance. An extensive clinical dataset with a total of 10 medical indicators was used, and SMOTE was used. The hybrid model achieved a high accuracy of 96.8% and outperformed standalone models such as SVM and Random Forest (with accuracy of 94.8% and 94.6% respectively). The model also achieved a recall of 1.00 for CKD cases, indicating that all CKD patients would be identified. It is worth noting that despite achieving best performance, the model still has a significant computational load, suggesting that more optimization is required for clinical application. The work marked a significant step forward in CKD classification by combining deep learning and classical machine learning. It was found that the combination of CNN and SVM was good at overcoming overfitting problems. Meanwhile, reducing the number of features using CNN before feeding them into SVM could also help mitigate the class imbalance problem. The conclusions of the study indicated that the hybrid model could be considered as a robust clinical application that can be further studied for future research.