I tried to make a logistic regression classifier to predict the class of unseen data. In this data cross terms play important role to predict the multi imbalance class, with this information I cannot achieve micro/weighted/macro precision/recall/F1 score more than 50% in any case. I believe that this is due to complex nature of the data. Then I read many questions about multicolinearity and orthogonal polynomial specially in R. I don't have any adavanced training in ML/statistics so I just want to make sure that removing the multicolinearity or generting the orthogonal polynomial would not change the prediction accuracy of the model. If prediction accuracy going to be increase then how should I incorportate in my model. Thanks.
from sklearn.preprocessing import PolynomialFeatures, StandardScaler log_reg_model = LogisticRegression(max_iter=50000,penalty='l2',multi_class='multinomial',class_weight='balanced',solver='lbfgs') pipe=Pipeline([('polynomial_features',polynomial),('StandardScaler',StandardScaler()), ('logistic_regression',log_reg_model)]) pipe.fit(x_train, y_train)