This study introduces a machine learning platform that classifies cancer cell phenotypes based on single-cell particle uptake profiles. Three human cancer cell line pairs—H460 (cisplatin-sensitive vs. resistant), PC3M (primary vs. metastatic), and A375 (low vs. high uptake)—were exposed to fluorescent polystyrene particles of five sizes (0.04 to 3.36 μm), and uptake was quantified via flow cytometry. Morphological features alone could not reliably distinguish subtypes, especially in PC3M and A375 cells. In contrast, models incorporating uptake data, particularly granularity (SSC), achieved classification accuracies exceeding 95% in cell-grouped analysis using random forest, SVM, and XGBoost. Single-cell models were less accurate, but combining uptake with physiological features improved performance. Uptake correlated more strongly with granularity than size, and its inclusion helped normalize inter-experiment variation. Grouped analysis with 500-cell batches outperformed single-cell analysis, showing reduced batch effects and greater reproducibility. Classification accuracy declined with smaller group sizes, especially in A375 cells, and the largest particles (3.36 μm) performed worst. PCA and feature binning partially recovered performance in independent batch experiments. Particle uptake generated functional biophysical fingerprints that allowed classification of cancer cells with similar morphology but distinct behavior, demonstrating that cellular internalization capacity reflects phenotypic differences in drug resistance and malignancy.