MNDO

Python implementation of MNDO (Multivariate Normal Distribution based Oversampling).

Requirements

If you use Keel-datasets, you can use the following command.

python pre_dataset.py dataset_directory

Preprocessing all files in a directory.
Remove unnecessary lines and replace class labels. (Positive class -> 1, Negative class -> 0)
Preprocessed data is saved in MNDO/Predataset/xxx.csv

Resampled(generated) data is stored in ./pos_data

python over-sampling.py data_path

python train.py data_path

train.py steps:

Load data
Over-sampling (MNDO, SMOTE, Borderline-SMOTE, ADASYN, SMOTE-ENN and SMOTE-Tomek Links)
Scaling (Normalization or Standardization)
Learning (SVM, Decision Tree and k-NN)
Predict (Results is saved in MNDO/output/xxx.csv)

If you want to train all files, you can use this script:

./run.sh

Kotaro Ambai, Hamido Fujita, MNDO: Multivariate Normal Distribution Based Over-Sampling for Binary Classification, Volume 303: New Trends in Intelligent Software Methodologies, Tools and Techniques, DOI: 10.3233/978-1-61499-900-3-425
Study on improving prediction accuracy for imbalanced medical data using Multivariate Normal Distribution based Oversampling

Kotaro Ambai (baibai25)