The fabrication of YBa₂Cu₃O_7–δ (YBCO) high-temperature superconducting thin films depends on numerous processing parameters; however, acquiring large-scale experimental data during actual film growth remains challenging. This study utilizes a small-sample dataset of ten parameters (substrate type, substrate size, laser focal length, target-substrate distance, output voltage, laser frequency, oxygen pressure, growth time, substrate temperature, and annealing time) as inputs to examine the guiding role of a multilayer perceptron (MLP) neural network in optimizing pulsed laser deposition (PLD) processes. The outputs are the (005) peak position and the full width at half maximum (FWHM) of the rocking curve obtained from X-ray diffraction (XRD). The MLP incorporates bootstrap resampling and early stopping to mitigate overfitting under small-sample conditions. The model gives reasonable predictions and favorable generalization performance, validating the feasibility of data-driven approaches in modeling complex thin-film processes. Notably, the root-mean-square error (RMSE) for peak position is 0.01°, approaching the instrumental resolution of XRD measurements (±0.005°). Furthermore, Pearson correlation analysis explores the interrelations: the peak position (lattice constant) is primarily influenced by oxygen content and epitaxial stress, while the FWHM (crystalline quality) depends more sensitively on complex kinetics and exhibits strong nonlinearity. This work establishes a scalable and transferable analytical framework for superconducting thin-film experiments under small-sample constraints, providing a feasible pathway for data-driven process optimization and facilitating in-depth interpretation of process-structure-property correlations. The datasets presented in this paper are openly available at https://doi.org/10.57760/sciencedb.00zs1.