In addition to this and considering one of the criteria

"1. Model accuracy: Accuracy evaluation of the landslide susceptibility map using validation samples, with consideration of the balance between the training dataset and classification accuracies",

the training samples are polygon while the validation samples are point geometry. My approach entailed extracting points within the polygons to have that geometry match, but I am having miliions of trainig points while the validation points are just 40,000.

Will be this creteria be based on the initially provided train polygons or the number of points i generated from this for my case?