Chronic Kidney Disease (CKD) refers to a persistent and progressive impairment of kidney function occurring over a prolonged duration. Impaired kidney filtration can lead to the accumulation of waste products and excess fluid in the bloodstream, contributing to the development of secondary medical conditions. CKD leads to high blood pressure, glomerulonephritis, diabetes, and polycystic kidney disease. However, early detection of CKD is significant for decreasing complications and preventing kidney failure. However, generalization and class imbalance issues complicate the detection process. In order to improve CKD detection and resolve current limitations, an optimized deep learning approach is presented in this paper. This paper proposes a CKD detection framework that integrates XceptionNet with the Harmonic Addax Optimization Algorithm (HAOA). First, the chronic kidney dataset is provided as input and undergoes sigmoid normalization to ensure proper data scaling and structural consistency. Next, feature fusion is performed by a Deep Belief Network (DBN) with a Soergel metric. Then, data augmentation is performed utilizing the Synthetic Minority Overlapping Technique (SMOTE). At last, CKD detection is done using Xception with HAOA. Here, HAOA is developed by combining Harmonic analysis and the Addax Optimization Algorithm (AOA). The performance of the proposed Xception with the HAOA method is analyzed by the CKD dataset 1, CKD dataset 2, and the Risk Factor Prediction of CKD Dataset. It also achieves a good True Positive Rate (TPR) value of 94.679 %, True Negative Rate (TNR) of 92.777 %, and accuracy of 93.667 %, a precision of 92.258 %, and an F1-score of 93.453 %. The proposed model serves as an effective tool for early CKD diagnosis, reducing the risk of kidney failure and improving potential outcomes.