We report on a new method for the analysis of hot-film anemometry in turbulent bubbly flows based on stochastic pattern recognition summarizing our results of [J.M. Rensen, S. Luther, J. de Vries, D. Lohse, Hot-film anemometry in bubbly flow I: bubble–probe interaction, in press; S. Luther, J. Rensen, Hot-film anemometry in bubbly flow II: local phase discrimination, in press; J.M. Rensen, S. Luther, D. Lohse, Velocity structure functions in turbulent two-phase flows, J. Fluid Mech., in press]. It consists of an optimal signal decomposition using an adaptive wavelet transform and neural network based classification. We discuss the application of autoregressive models to obtain energy spectra for gapped time series.