Classification of Gaussian random processes using a simple linear discriminant and its application to seismic differentiation (Q5966493)

The problem of discriminating between two Gaussian processes observed through time series is tackled as caused by earthquakes and underground nuclear explosions. To avoid the complication of likelihood ratio methodology the author proposes to use an optimal linear discriminant function, which amounts to identifying an optimal vector \(\ell_ 0\) chosen in such a way as to minimize the maximum error of misclassification. It is shown how \(\ell_ 0\) may be identified and that the procedure is equivalent to the estimation of the multiplier in a weighted least squares model. The further simplification is then introduced by using a simple least squares estimator, and the consequences of doing so on the efficiency of the procedure are carefully analysed. In conclusion the method is adapted and applied to the discrimination between earthquakes and underground nuclear explosions for seven events and shown that in each case the event is correctly classified.