Overview

JBoost is an easy to use and modify tool for boosting classification. JBoost includes state-of-the-art algorithms and can be used by researchers to quickly implement new boosting algorithms. JBoost also includes a set of easy to use scripts so that machine learning novices can quickly learn and utilize the power of boosting.

Some of the algorithms currently implemented include AdaBoost, LogitBoost, BrownBoost, and BoosTexter. These algorithms are wrapped inside of an implementation of alternating decision trees (ADTrees), which allows for easy visualization of the final classifier, even for high dimensional data. Each of the algorithms comes with a set of options that allows for customization to your dataset.

To learn more, download JBoost or read the documentation.

Power of JBoost

AdaBoost, the original adaptive boosting algorithm, is one of the simplest algorithms in machine learning. An experienced programmer could likely code an implementation in ten minutes. However, JBoost is much more than AdaBoost. JBoost has several (soon to be more!) learning algorithms with easy to use, standardized, customizable, tools.

Some of the features include:

• Multiple boosting algorithms. JBoost can utilize the power of AdaBoost, exploit the noise resistant properties of BrownBoost, or asymmetric cost of NormalBoost (soon to come!).
• Visualization. Original boosting methods focused on combining many decision trees, causing the final combined classifier to have millions of nodes. ADTrees have the same expressive power, while containing orders of magnitude fewer nodes. This leads to robust classifiers that are easily understandable and human editable.
• Multi-Label. Many problems demand that each example can be classified in multiple ways. For instance, a movie may be "romantic" and a "comedy".
• Multi-Class. Many classification algorithms (including AdaBoost) were originally designed to classify two classes (for instance, is the car "broken" or "not broken"). JBoost can classify an arbitrary number of classes.

Other Projects

If JBoost doesn't provide sufficient classification, we encourage the user to read the tips section. However, JBoost isn't the perfect tool for all purposes. Here is a very incomplete list of other projects with similar goals:

• WEKA - A general purpose data mining tool that is equipped with many types of classifiers.
• SVM^light - A support vector machine package with many kernel options.