Strojno učenje

1 Priprava podatkov

Izbrane spremenljivke (p < 0.10):

Target: DSWI01 | Predictors: 53

Izvedeno skaliranje s standardizacijo numeričnih spremenljivk in “one-hot encoding” za kategorijske spremenljivke.

CV results:

Model	CV Accuracy	CV F1	CV ROC AUC
LogisticRegression	0.913 [0.906, 0.920]	0.444 [0.397, 0.522]	0.855 [0.814, 0.866]
DecisionTree	0.894 [0.877, 0.901]	0.514 [0.471, 0.552]	0.731 [0.716, 0.759]
RandomForest	0.934 [0.927, 0.938]	0.555 [0.478, 0.603]	0.943 [0.939, 0.960]
GradientBoosting	0.943 [0.929, 0.945]	0.662 [0.551, 0.676]	0.929 [0.885, 0.937]
AdaBoost	0.893 [0.892, 0.893]	0.000 [0.000, 0.000]	0.811 [0.783, 0.852]
XGBoost	0.950 [0.945, 0.958]	0.706 [0.672, 0.761]	0.941 [0.909, 0.950]
LightGBM	0.953 [0.949, 0.963]	0.729 [0.692, 0.792]	0.944 [0.920, 0.963]
SVC	0.933 [0.925, 0.940]	0.550 [0.483, 0.613]	0.894 [0.891, 0.934]
KNN	0.928 [0.925, 0.930]	0.583 [0.567, 0.606]	0.858 [0.847, 0.914]
MLP	0.932 [0.917, 0.937]	0.645 [0.608, 0.695]	0.889 [0.836, 0.930]

CV setup: StratifiedKFold with 5 folds (shuffle=True, random_state=1974).

CV results:

Model	CV Accuracy	CV F1	CV ROC AUC
LogisticRegression	0.786 [0.763, 0.815]	0.423 [0.390, 0.484]	0.857 [0.818, 0.873]
DecisionTree	0.805 [0.767, 0.830]	0.433 [0.417, 0.458]	0.819 [0.785, 0.835]
RandomForest	0.930 [0.922, 0.935]	0.533 [0.422, 0.567]	0.945 [0.939, 0.962]
GradientBoosting	0.956 [0.952, 0.958]	0.759 [0.719, 0.770]	0.942 [0.913, 0.958]
AdaBoost	0.897 [0.893, 0.908]	0.067 [0.065, 0.245]	0.852 [0.806, 0.880]
XGBoost	0.955 [0.950, 0.957]	0.743 [0.710, 0.759]	0.950 [0.918, 0.961]
LightGBM	0.958 [0.953, 0.965]	0.757 [0.729, 0.813]	0.942 [0.909, 0.955]
SVC	0.940 [0.937, 0.950]	0.647 [0.611, 0.710]	0.925 [0.900, 0.932]
KNN	0.934 [0.932, 0.940]	0.602 [0.574, 0.657]	0.910 [0.882, 0.927]
MLP	0.932 [0.917, 0.937]	0.645 [0.608, 0.695]	0.889 [0.836, 0.930]

CV setup: StratifiedKFold with 5 folds (shuffle=True, random_state=1974).

Shranjeni modeli: 10