The global demand for plant-based food products, including cheese analogues, is continuously increasing. However, most scientific attention has focused on their nutritional value rather than on functional properties such as melt-stretch. This study integrated the rheological features (pasting and viscoelasticity) of nine different starches with machine learning analysis to identify melt-stretchable starches. These starch samples were further evaluated in the formulation of plant-based cheese analogues. Hydroxypropyl tapioca, waxy potato, and tapioca starches showed the highest melt-stretch length values, compared to the other starch samples. The melt-stretch properties of the starch samples did not show distinct correlations with their individual pasting parameters, while they had lower viscoelasticity as well as high frequney dependence. FTIR analysis demonstrated that the decreased ratio of crystalline to amorphous regions upon conversion to a paste state contributed to the melt-stretch properties of starches. When the pasting and viscoelastic results were subjected to machine learning analysis, the starch samples were well-clustered based on their melt-stretchability. Furthermore, the starch samples with good melt-stretchability (hydroxypropyl tapioca, waxy potato, and tapioca starches) were successfully identified by the logistic binary classification model trained using rheological datasets, specifically, dynamic viscoelasticity. Similar melt-stretch features were moreover observed in the starch-incorporated plant-based cheese analogues, confirmed by a distinct decrease in the storage moduli over temperature and the highest values of melting index.