Ensembling Classifiers - An application to image data classification from Cherenkov telescope experiment

Bagging have been in active research and shown improvements in classification results for several benchmarking data sets with mainly decision trees as their base classifiers. In this paper we experiment to apply these Meta learning techniques with classifiers such as random forests, neural networks and support vector machines. The data sets are from MAGIC, a Cherenkov telescope experiment. The task is to classify gamma signals from overwhelmingly hadron and muon signals representing a rare class classification problem. We compare the individual classifiers with their ensemble counterparts and discuss the results. WEKA a wonderful tool for machine learning has been used for making the experiments. I. PROBLEM DOMAIN AGIC [1] is the Cherenkov telescope used to detect the gamma rays from the outer universe. It is designed to provide vital information on several established gamma-ray sources, like Active Galactic Nuclei, Supernova Remnants, Gamma Ray Bursts and Pulsars. It collects gamma ray events (in little quantities) along with many other particle events represented as images shown in the (Fig.1). The pixels making up the image can be converted to some set of image parameters also called as hillas parameters by various image processing and feature extraction techniques [2], which statistically allow a separation of events. A gamma ray signal defines an ellipse in the camera plane of the telescope where as the other showers make up an error ellipse plane (Fig.1). The data sets used in the experiment contain 10 image parameters. Due to atmospheric radiations, the ground based telescope collects overwhelming events of hadrons and muons also called as background. To understand the gamma ray sources, it is an important task for separating gammas from other particles. There is only a weak discrimination between the gamma and background events, making the data an excellent proving ground for the classification techniques [2]. Things are further complicated by added noise in the data collection by the hardware bias in the telescope. As the sources of high-energy gammas are few and their signal comparatively weak, creating a case for the rare class classification problem. It will be a daunting task to separate the gamma signals from these overwhelming sea of background signals, both having very similar characteristics.. Classification of signals plays a vital role for making the astronomical analysis of gamma ray objects, and any small improvements in the classification accuracy will be significant in these analysis tasks. Automated classification of vector data into …