Traditional drug discovery is time-intensive and costly, often spanning over a decade and incurring billions in expenses. This study introduces a novel machine learning pipeline tailored to predict and optimize inhibitors for Enhancer of Zeste Homolog 2 (EZH2), a critical epigenetic target implicated in cancer progression. Leveraging curated datasets from repositories like the Protein Data Bank, PubChem, and ChEMBL, the pipeline integrates feature selection using Lipinski's Rule of Five with advanced regression algorithms, achieving predictive metrics of R² = 0.75 and RMSE = 0.8 for inhibitory potency (pIC50 values). These results highlight the pipeline's strong predictive accuracy and reliability in identifying potent inhibitors. Unique to this approach is the focus on biologically interpretable descriptors, such as molecular weight and LogP, which enhance model transparency and relevance to pharmacokinetics. Validation through molecular docking (SwissDock) and RDKit reinforced robustness, with the model demonstrating a threefold improvement in efficiency by narrowing chemical libraries and reducing experimental burdens. By combining machine learning with pharmacological insights, this study addresses key bottlenecks in early-stage drug discovery, providing a scalable and adaptable framework for EZH2-targeted cancer therapeutics. While experimental validation remains indispensable, this computational approach significantly accelerates the prioritization of candidate compounds, contributing to cost-effective and efficient oncological drug development.