Techniques based upon boosting to counter class imbalance problem — A survey
Prabhjot Kaur, Vasu Negi · International Conference on Computing for Sustainable Global Development · 2016
Today every data mining application suffer from class imbalanced problem. A precipitous rise in the interest was noticed amongst the machine learning enthusiasts. The pertinent reason for such an attention is that it is one of the reasons that degrade the performance of classifiers. The problem arises when one class (minority) has significantly fewer number of instances when compared to the number of instances of other class (majority). As a consequence, when traditional classification algorithms are applied they show their biased behavior towards the majority class, thus forming a model that may be accurate but often not that useful. Several techniques have been developed to counter the problems related with class imbalance problem. Boosting-an ensemble based learning technique has proved its cardinality in the improvement of the prediction of the minority by countering the effect biased behavior of the classifiers against the class with the fewer number of instances. In this work, we survey different techniques that involve Boosting as their combination scheme.