Effect of Dataset Size and Train/Test Split Ratios in QSAR/QSPR Multiclass Classification

Applied datasets can vary from a few hundred to thousands of samples in typical quantitative structure-activity/property (QSAR/QSPR) relationships and classification. However, the size train/test split ratios greatly affect outcome models, thus classification performance itself. We compared several combinations dataset sizes with five different machine learning algorithms find differences or similarities select best parameter settings nonbinary (multiclass) It is also known that models are ranked differently according merit(s) used. Here, 25 parameters were calculated for each model, then factorial ANOVA was applied compare results. The results clearly show not just between but lesser extent ratios. XGBoost algorithm could outperform others, even multiclass modeling. reacted change sample set size; some them much more sensitive this factor than others. Moreover, significant be detected as well, exerting great effect on test validation our models.