In this paper a statistical model for noisy data selection has been presented. It combines two powerful tools: a local wavelet analysis and multidimensional data analysis of wavelet transform coefficients. In the proposed model the adapted Malvar wavelet transform has been applied. It leads to a partition of the measuring signal to isolate transients. The multidimensional wavelet coefficients analysis has been applied to constitute a set of discriminating parameters that can be used to explore features characterizing transients caused by the air bubbles from diver's oxygen tanks.