In contemporary cosmology, precise data analysis is essential for understanding the universe’s most profound mysteries. Upcoming surveys will provide unprecedented high-precision data, offering new insights into “dark” aspects of the cosmological model, such as the nature and evolution of Dark Matter and Dark Energy. Yet, tensions within the standard ΛCDM model—such as discrepancies in the Universe’s expansion rate (Hubble tension) and the amplitude of clustering—underscore the need for innovative statistical techniques. This talk introduces two novel methods to address these challenges. Bayesian Model Averaging (BMA) provides a principled framework to incorporate model uncertainty as an additional source of error while enabling data-driven model selection. We demonstrate its application to the Hubble tension, exploring Early Dark Energy as an alternative to ΛCDM and analyzing the most prominent extensions of the standard model using the latest CMB, LSS and SNIa datasets. The second focus is on cosmological parameter biases in LSS analyses, particularly the projection effects arising when extending analyses to small-scale features. These biases stem from nuisance parameter priors, which can feedback into and distort linear-regime cosmological parameters. This talk presents an innovative solution: a reparameterization of nuisance parameters using non-linear transformations via Generalized Additive Models (GAMs), effectively mitigating projection effects. By leveraging well-established techniques from data-intensive fields, this talk demonstrates how these approaches unlock the full potential of cosmological datasets, advancing the cosmological model and overcoming longstanding limitations in standard analyses.