Semi-supervised Adaptive Discriminative Discretization (SADD) for Naive Bayes classifier

SADD is a semi-supervised discretization framework that can be universally applied in various naive Bayes classifiers. There are two main stages in SADD:

• Pseudo labeling

  The k-nearst-neighboors is used to derive the pseudo labels for unlabeled data. The default k is set as 1.

• Adaptive dicriminative discretization

  An adaptive threshold is designed to derive the discretization scheme with less information loss.

Implementation

This is an source code of SADD for naive Bayes classifier by using MATLAB. The version of MATLAB should be >=2019b.

• To successfully run the code, you need install the Bioinformatics Toolbox and Statistics and Machine Learning Toolbox.

• Dowload the source file, place it MATLAB and then run the main.m file.

Dataset

We provide a zip file IndoorLoc.zip for one of the used datasets in the paper and it contains a CSV file IndoorLoc.csv and read.me for dataset description. The example dataset IndoorLoc.csv have total 21048 samples, 520 features with 3 classes. The detailed descriptions of all datasets used in the paper can be found in https://archive.ics.uci.edu/ml/index.php.

All the features should be first convert to numerical value, and then make a classification by NB classifiers.

If you use this code, please cite:

@article{wang2023semi,
  title={A semi-supervised adaptive discriminative discretization method improving discrimination power of regularized naive Bayes},
  author={Wang, Shihe and Ren, Jianfeng and Bai, Ruibin},
  journal={Expert Systems with Applications},
  volume={225},
  pages={120094},
  year={2023},
  publisher={Elsevier}
}