Lightcurve Classification in Massive Variability Surveys II: Transients towards the Large Magellanic Cloud

Abstract

Automatic classification of variability is now possible with tools like neural networks. Here, we present two neural networks for the identification of microlensing events -- the first discriminates against variable stars and the second against supernovae. The inputs to the networks include parameters describing the shape and the size of the lightcurve, together with colour of the event. The network computes the posterior probability of microlensing, together with an estimate of the likely error. An algorithm is devised for direct calculation of the microlensing rate from the output of the neural networks. We present a new analysis of the microlensing candidates towards the Large Magellanic Cloud (LMC). The neural networks confirm the microlensing nature of only 7 of the possible 17 events identified by the MACHO experiment. This suggests that earlier estimates of the microlensing optical depth towards the LMC may have been overestimated. A smaller number of events is consistent with the assumption that all the microlensing events are caused by the known stellar populations in the outer Galaxy/LMC.