A NEW CENTROID BASED ALGORITHM FOR HIGH SPEED BINARY CLASSIFICATION
Binary classification is a popular research topic with many well-known algorithms available including Support Vector Machines, neural networks, Bayesian networks and decision trees. These algorithms are aimed at high accuracy, many times at the cost of training speed. We present a new binary classification algorithm which determines the separating hyperplane’s maximized margins to classify the data. However, instead of the analytic approach of the existing Support Vector Machine algorithms, we approach the problem using a new centroid based training method. This method alternates between heuristic approximations of the weighted centroid and the maximization of the margins of the separating orthogonal hyperplane. This algorithm provides accuracy close to other well-known methods while gaining significant speed increases. We present future directions for research including as a precursor to other well-known algorithms.
School Location:USA - Ohio
Source Type:Master's Thesis
Keywords:binary classification centroid maximized margins svm
Date of Publication:01/01/2004