Performance of non-parametric reconstruction techniques in the late-time universe

In the context of a Hubble tension problem that is growing in its statistical significance, we reconsider effectiveness non-parametric reconstruction techniques which are independent prescriptive cosmological models. By taking cosmic chronometers, Type Ia Supernovae and baryonic acoustic oscillation data, compare contrast two important approaches, namely Gaussian processes (GP) \textbf{Lo}cally w\textbf{e}ighted \textbf{S}catterplot \textbf{S}moothing together with \textbf{Sim}ulation \textbf{ex}trapolation method (LOESS-Simex or LS). these methods, besides not requiring model, they also do require physical parameters their approach to data (but depend on hyperparameters). We firstly show how both GP LOESS-Simex can be used successively reconstruct various sets high level precision. then directly approaches quantitative manner by considering several factors, such as well reconstructions approximate themselves respective uncertainties evolve. light puzzling tension, it consider uncertain regions evolve over redshift methods for estimating at current times. For chronometers compiled sets, find generically produce smaller variances reconstructed minimum value $\sigma_{\rm GP-min} = 1.1$, while situation LS totally different LS-min} 50.8$. Moreover, some characteristics alleviate low $z$, where presents less underestimation comparison GP.