Some machine learning algorithm for predict the 20 news groups dataset. It uses the Weka API to generate classifiers and display some statistics, the confustion matrix and the ROC curve for some experiments.

You need to have Java 8 JRE or JDK installed in your computer with JAVA_HOME set in your env variables.

You can execute the experiments running bin/newsgroup-classication-exe [OPTIONS]

The options are:

--adaboost-multinomial-experiment

This experiment create a classifier using AdaBoost algorithm with Naive Bayes Multinomial. You can use the parameters -number-of-words to specify how many words will be used in the vocabulary. -tranning-size specifies the percentage of instances that will be used in tranning.

--arff-experiment

This experiment export the ARFF training and testing files

--dataset-dir

Set the dataset directory path

--decision-tree-experiment

This experiment create a classifier using decision-tree algorithm. You can use the parameters -number-of-words to specify how many words will be used in the vocabulary. -tranning-size specifies the percentage of instances that will be used in tranning.

--download-dataset

This option will download the dataset from internet

--find-best-number-of-attributes-experiment

This experiment evaluate classifiers with diferent attribute number trainning. It uses Naive Bayes and Naive Bayes Multinomial to build the classifiers. By default experiment will run with 100, 1000, 2500, 5000, 20000, 40000 words

--help

Print this message

--knn-experiment

This experiment create a classifier using knn algorithm. You can use the parameters -number-of-words to specify how many words will be used in the vocabulary. -tranning-size specifies the percentage of instances that will be used in tranning.

--naive-bayes-experiment

This experiment create a classifier using naive-bayes algorithm. You can use the parameters -number-of-words to specify how many words will be used in the vocabulary. -tranning-size specifies the percentage of instances that will be used in tranning.

--naive-bayes-multinomial-experiment This experiment create a classifier using naive-bayes multinomial algorithm. You can use the parameters -number-of-words to specify how many words will be used in the vocabulary. -tranning-size specifies the percentage of instances that will be used in tranning.

bin/newsgroup-classication-exe --download-dataset --naive-bayes-experiment

To build and package the distribution just execute ./gradlew clean build distZip. It will generate a zip file in build/distribution directory.