Inter-cell voltage imbalance degrades efficiency, accelerates aging, and increases failure risk in electrochemical energy storage systems. This study models and predicts balancing-charger conditions using two machine-learning algorithms Random Forest (RF) and K-Nearest Neighbors (KNN) across packs of 4, 8, 10, and 15 cells with five dataset scales (1,000; 5,000; 10,000; 15,000; and 20,000 samples). Voltage data were obtained through simulation and laboratory measurements on lithium-ion cells within 3.2–4.2 V, then normalized and split into training and testing sets. Performance was evaluated using accuracy, confusion matrices, and feature-importance analysis. Results show RF achieves 0.98 accuracy for 4-cell packs and remains high at 0.93 for 15-cell packs, whereas KNN attains only 0.94 and 0.37 on the same configurations. RF exhibits predictions concentrated along the confusion-matrix diagonal with well-distributed feature weights, indicating robustness to increasing dimensionality. The contributions are threefold: (1) an evaluation framework for comparing classifiers in multi-cell scenarios; (2) empirical evidence of RF’s scalability for detecting balancing conditions from single-voltage inputs; and (3) practical implications for BMS operation more accurate balancing decisions, prioritization of problematic cells, reduced futile equalization cycles, and potential energy savings together with extended service life. These findings recommend RF as a core algorithm for machine-learning-based balancing chargers, particularly for real-world deployment on power-constrained edge devices.