Statistical Methods
SPSS 26.0 and Modeler 18.0 (IBM Corporation, Armonk, New York, United
States) programs were used to analyze variables. Univariate data’s
suitability to normal distribution was evaluated with the Shapiro-Wilk
Francia test, while variance homogeneity was assessed with the Levene
test. Independent-Samples T-test was used together with Bootstrap
results. In contrast, the Mann-Whitney U test was used together with
Monte Carlo to compare two independent groups according to quantitative
data. One-Way Anova test, one of the parametric methods, was used to
compare multiple separate groups according to quantitative data. The
Tukey HSD test was used for post hoc analysis. Kruskal-Wallis H Test,
one of the nonparametric tests, was used with Monte Carlo simulation
technique results, and Dunn’s test was used for Post Hoc analysis. In
comparison of categorical variables, Pearson Chi-Square Exact results
were analyzed, while the Fisher-Freeman-Holton test was tested with the
Monte Carlo Simulation technique.
Logistic Regression, Support Vector Machine, Random Forest, K-nearest
Neighbor Algorithm, Simple (Native) Bayes Classification, and Neural
Network (Multilayer Perceptron-Radial Basis) were used to find and
predict the variable with the highest significance in the patient and
control groups. Neural Network (Multilayer Perceptron) analysis, which
is the most successful model among these methods, was used. Gradient
descent was used for optimization algorithm, Hyperbolic tangent as
hidden layer activation function, Softmax as output Layer activation
function were used. The Mini-Batch method was used for the training data
selection, and a 70% Trial set was set as a 30% Testing set.
Quantitative variables are mean ± SD (standard deviation) in tables.
Moreover, Median (Percentile 25% / Percentile 75%), while categorical
variables were shown as n (%). Variables were analyzed at a 95%
confidence level, and a p-value of less than 0.05 was considered
significant.