The Dirichlet process is an important fundamental object in nonparametric Bayesian modelling, applied to a wide range of problems in machine learning, statistics, and bioinformatics, among other fields. This flexible stochastic process models rich data structures with unknown or evolving number of clusters. It is a valuable tool for encoding the true complexity of real-world data in computer models. Our results show that the Dirichlet process improves, both in distribution density and in signal-to-noise ratio, with larger sample size; achieves slow decay rate to its base distribution; has improved convergence and stability; and thrives with a Gaussian base distribution, which is much better than the Gamma distribution. The performance depends greatly on the choice of base distribution. The higher the value of α (a concentration parameter), the better the clustering and noise suppression. The distributional behavior of data can be approximated rigorously by the biorthogonal wavelet analysis. Since the Dirichlet process is an interesting object of observation, we computed it for a few wavelet bases and among them, we found that the Cohen-Daubechies-Feauveau (CDF) basis is the one that captures the Dirichlet process most accurately. Our results may be useful in applying the Dirichlet process to real-world experimental data and in developing Bayesian non-parametric methods.