An Enchanced Approach for Early Chronic Kidney Disease Prediction Using Advanced Machine Learning and Data Processing Techniques

Chronic renal disease is a medical disorder in which the kidneys gradually slow down, reducing their ability to filter blood. Diabetes, high blood pressure, heart disease, and a family history of renal failure are the primary risk factors for this condition. The prognosis for patients with chronic kidney disease (CKD) varies according to the stage at which they are diagnosed. Stages 0 and 1 represent the early stages of the disease, whereas stage 4 represents the severe stages, where the kidneys are nearly at the point of failure. Early diagnosis is therefore essential for the patient's survival and the appropriate therapies. Blood tests, biopsies, urine tests, and imaging tests are among the medical diagnostic techniques that can be used to diagnose chronic kidney disease (CKD). However, these medical diagnostic techniques have certain drawbacks and restrictions. They are employed to identify diseases in their later stages and can occasionally yield erroneous results; they are also expensive. Consequently, the researcher has discovered a novel technique for the early identification of CKD that use machine learning algorithms to overcome these complications because machine learning algorithms are becoming increasingly significant in health care datasets. These methods also have issues with overfitting, class imbalance, biased results, problems with dataset interpretability, and high computing costs. Therefore, improved preprocessing techniques and machine learning algorithms are used in this research work to overcome the inadequacies of the existing approaches for CKD prediction. The model is tested using the Kaggle dataset on chronic renal disease. Class imbalance and missing attribute values are issues with this dataset. Additionally, this dataset has several categorical attribute values and additional variances in attribute values. A few preprocessing techniques are used to improve the data quality, including K-Nearest Neighbour for filling in missing attribute values, SMOTE for class imbalance, Label encoding for transferring categorical values, and Min-max scaling. The Adaptive Neural Optimized Model is used for the classification process, which leverages the Extreme Learning Model's power for classification. The particle swarm optimization method is used for feature selection and optimization. The algorithm's output is evaluated using a number of performance criteria, including as accuracy, precision, and recall. With the highest accuracy of 98%, the suggested model performs better than the current models.