A NEW MULTIVARIATE TECHNIQUE FOR TOP-QUARK SEARCH

We present a new multivariate event classifier, the PDE classifier, wh ich can be used to identify rare signals in the presence of large back grounds. The classifier is based on a discriminant function which uses kernel density estimates of the signal and background event densities . The PDE method offers the flexibility of a neural network but is con ceptually simpler, theoretically better understood, and easier to opti mize in practice. The technique is tested on Monte Carlo data using pr ocesses t ($) over bar t --> e + E(t) + jets as signal and W + jets -- > e + E(t) + jets as background. The performance of the PDE classifier is similar to that of a neural network. Both methods outperform conve ntional analysis.