Reliable prediction of extended in vitro drug release for long-acting injectable (LAI) microparticles formulated with poly(lactic-co-glycolic acid) (PLGA) is required to support formulation development in its early stages. However, conventional optimization relies on time- and resource-consuming trial-and-error studies and experimentation. To overcome these issues, a structured machine-learning (ML) framework was implemented to predict cumulative drug release after 24 h using formulation descriptors and early in vitro release data. A harmonized dataset of 347 PLGA formulations and 5265 individual release measurements were obtained from 2 published datasets. Multiple algorithms were evaluated using repeated group-based cross-validation under defined information ablations. Ensemble tree-based models consistently outperformed neural network, single-tree, and linear model approaches. Light Gradient Boosting Machine (LightGBM) scored the best performance with a mean absolute error of (MAE = 0.129), with relative accuracy observed for Random Forest and CatBoost models. The utilization of early-release features and kinetic parameters improved accuracy in comparison to descriptor-only models. SHAP-based interpretability analysis identified Time, Korsmeyer-Peppas release rate constant (k), and the surfactant percentage as dominant contributors. This work provides a unique methodological contribution using the structured predictive framework adopted by confining model predictions to time points occurring after the initial 24 h experimental interval which emulates early-stage formulation screening scenarios and prevents spurious enhancement of the predictive performance. The developed model predicts and reconstructs full release trajectories while maintaining calibration and minimal systematic bias. These findings demonstrate that interpretable ensemble learning, combined with meaningful mechanistic descriptors, can support predictive, reproducible, Quality-by-design (QbD) aligned optimization of PLGA-based long-acting injectable formulations. An online simulation tool for the predictive model is available in the following GitHub repository under the README section: Pharmaceutical-ML-lab/Predictive-Modelling-of-Drug-Release-from-PLGA-Microparticles-Using-AI.