Electrocatalytic water splitting offers a promising path toward sustainable hydrogen production, yet the oxygen evolution reaction (OER) remains a major bottleneck due to its inherently slow kinetics. While layered double hydroxides (LDHs) have emerged as effective non-precious-metal electrocatalysts, optimizing their multicomponent compositions is difficult given the vast number of compositional and synthetic parameters involved. To address this, an AdaBoost-based ensemble regression model (ABR) was built to predict OER overpotential at 10 mA·cm-2 using compositional descriptors, synthesis routes, and electrochemical parameters as inputs. The model enabled efficient screening of candidate LDH compositions and facilitated the identification of promising catalyst designs. Shapley Additive Explanations (SHAP) were then applied to pinpoint the compositional and synthetic factors most responsible for catalytic performance and to provide insights into the relative importance of different descriptors. Despite the small tabular dataset, the ABR model demonstrated satisfactory performance (MAE ≈ 33 mV) and recommended a nickel foam substrate, hydrothermal synthesis, and optimized LDH molar fractions. Based on these findings, a Zr-doped NiCo LDH (Ni0.75Co0.20Zr0.05) was engineered, achieving an overpotential of 234.5 mV at 10 mA·cm-2 along with a Tafel slope of 100.3 mV·dec-1. The experimentally measured performance confirmed the effectiveness of the machine learning-guided design strategy. Together, this machine learning-experimental strategy provides a systematic pathway for optimizing LDH-based electrocatalysts in sustainable energy conversion applications.