A comparative study and reference implementation of seven density-aware k-nearest-neighbors variants, plus vanilla kNN, on four UCI benchmark datasets.
This repository contains:
- Data preparation & preprocessing
- Implementations of:
- Basic kNN (from scikit-learn)
- Class‐Conditional Adaptive KDE kNN (Thesis Model)
- Local Distribution (Gaussian Model) kNN
- Density-Weighted kNN (Parzen-window)
- Region-of-Influence kNN
- Lift kNN
- Certainty-Factor kNN
- Experimental notebooks showing train/test splits, grid-search, runtime logging, metric calculation, and ROC-AUC plots
- Scripts for each algorithm under
scripts/so you can import and use them in your own projects - Results & visualization: heatmaps, barplots, ROC curves, run-time comparisons