Input Space

We have studied the application of diierent classiication algorithms in the analysis of simulated high energy physics data. Whereas Neural Network algorithms have become a standard tool for data analysis, the performance of other classiiers such as Support Vector Machines has not yet been tested in this environment. We chose two diierent problems to compare the performance of a Support Vector Machine and a Neural Net trained with back-propagation: tagging events of the type e + e ? ! cc and the identiication of muons produced in multihadronic e + e ? annihilation events. 1 Classiication algorithms Artiicial Neural Networks (ANN) are a useful tool to solve multi-dimensional classiication problems in high energy physics, in cases where one-dimensional cut techniques are not suucient. They are used both as hard-coded chip for very fast low-level pattern recognition in on-line triggering 1,2] and as a statistical tool for particle and event classiications in ooine data analysis. In ooine data analysis, a Monte Carlo simulation of the physics process and the detector response is necessary to train an ANN by supervised learning. ANN algorithms have been applied successfully in classiication problems such as gluon-jet tagging 3] and b-quark tagging 4]. The ANN classiiers 5] constructed in this paper have sigmoid nodes. Design and tuning issues were solved by applying practical experience rules 6].