Efficient Intrusion Detection Model Using Ensemble Methods

Abstract

Ensemble method or any combination model train multiple learners to solve the classification or regression problems, not by simply ordinary learning approaches that can able to construct one learner from training data rather construct a set of learners and combine them. Boosting algorithm is one of the most important recent developments in the area of classification methodology. Boosting belongs to a family of algorithms that has the capability to convert a group of weak learners to strong learners. Boosting works in a sequential manner by adding a classification algorithm to the next updated weight of the training samples by doing the majority voting technique of the sequence of classifiers. The boosting method combines the weak models to produce a powerful one and reduces the bias of the combined model. AdaBoost algorithm is the most influential algorithm that efficiently combines the weak learners to generate a strong classifier that could be able to classify a training data with better accuracy. AdaBoost differs from the current existing boosting methods in detection accuracy, error cost minimization, computational time and detection rate. Detection accuracy and computational cost are the two main metrics used to analyze the performance of AdaBoost classification algorithm. From the simulation result, it is evident that AdaBoost algorithm could able to achieve high detection accuracy with less computational time, and minimum cost compared to a single classifier. We have proposed a predictive model to classify normal class and attack class and an online inference engine is being imposed, either to allow or deny access to a network.