A Robust Machine Learning Pipeline for Chronic Kidney Disease Prediction using Outlier Detection, Feature Selection, and Adam Optimization

Data mining technology for healthcare purposes transforms basic medical information into useful insights which helps doctors predict diseases along with diagnosing patients while tailoring personalized treatment plans. A successful healthcare analytics process requires a complete data pipeline which includes data preprocessing followed by extraction then selection after optimization until classification. Data preprocessing implements the Z-Score Method together with Isolation Forest along with Interquartile Range (IQR) Method to detect outliers which would potentially harm model performance. The Principal Component Analysis (PCA) serves as an extraction method to reduce dimensions alongside log transformation which addresses data skewness to establish data stability. Wrapper methods Recursive Feature Elimination (RFE) together with Genetic Algorithms build predictive accuracy by continuously refining the selected feature subsets. Model performance enhancement relies on the Adam optimizer together with other optimization methods that aid in achieving high accuracy during medical outcome predictions. Adam Optimization optimization achieved a maximum training accuracy of 1.0 while validation accuracy reached 0.9 as the optimal value. Improvements in loss functionality decreased it from 45 to almost zero levels proving successful learning takes place. Analysis showed that the model demonstrated a positive trend in both validation accuracy levels despite periodic accuracy difficulties because the pipeline functioned well for predicting CKD conditions. Future work will direct its attention towards adjusting model performance by fine-tuning along with regularization techniques and additional data acquisition for enhancing generalization capabilities