ryanlacroix/binary-bayesian-classifiers

Performs bayesian classification on binary datasets using k-fold cross validation.

Binary Bayesian Classifiers

Python scripts for performing both independent and dependent Bayesian classification on binary datasts. Performance on randomly generated dependent datasets averages to ~90% correctness when used on datasets of 4 classes. Confusion matrices describing performance in more detail may be found in classifier directories.

Both dependent and independent classifier algorithms are trained and tested using a default 5-fold cross validation scheme. Number of folds can easily be modified in the k_fold_validator() parameters.

Usage

• Datasets based on randomly generated dependence trees can be created by running generate_data.py in the dependent_data_generator directory. Although probabilities are entirely random upon each class generation, the structure of the trees across all generated datasets may be modified in the dep_tree.py script.

• The classifier can be invoked by running the cross_validator.py script within the classifier directories. This script should be general enough to run classification on datasets of any class number, length and feature dimensions.

• The wine dataset can be converted to binary form by running process_wine_info.py from within the wine_dataset_interpreter directory. Output is ready to be fed into cross_validator.py for classification. A seperate directory is already set up for the wine classifier, containing all datasets and minor invocation differences already implemented.

Note

Running the classifier scripts requires both NetworkX and Matplotlib