EVALUATION OF MACHINE LEARNING HYPERPARAMETERS PERFORMANCE FOR MICE PROTEIN EXPRESSION DATA IN DIFFERENT SITUATIONS

In this study, the aim was to assess effect and significance of hyperparameters in four different datasets containing values for observation numbers variable counts with machine-learning methods support vector machines artificial neural networks. With aim, a dataset comprising 15 repeats 77 protein levels from 38 healthy 34 down syndrome mice used. A total 138 models model classification performance criteria were obtained study combinations methods. Comparison used like accurate percentage, kappa statistic, mean absolute error square root squares. According criteria, first 1080 observations x variables had 71.30% percentage assumed parameters polynomial kernel function, while changing hyperparameter increased rate 99.44%. Similarly, second 50.65% network single hidden layer 2 neuron model, 90.46%. conclusion, situations low numbers, machine learning determined display lower performance. However, datasets, it is very important networks machines, especially radial basis function functions, set according dataset. especially, gain importance.