a. Size of Training Set - (3910, 58)
b. Size of Testing Set - (691, 57)
c. The column of responses i.e., ['Y'] is binary.
d. There are 57 predictors, none of which have informative headers.
There is no significant Data Imbalance in the dataset.
a. There are no null values in any column (Missing data handling not required).
b. There is no column filled with all zeroes.
The dataset has a significant amount of noise which gives rise to a lot of outliers whose elimination is not necessary since “An Outlier is that observation which is significantly different from all other observations.” which is not the case.
To handle highly co-related features since two or more collinear features (correlated in some way but not necessarily with a strictly linear relationship) can give unexpected results while computing feature importance.
New Training Set Shape - (3910, 52)
Final Training Set Shape - (3910, 36)
Step 1: Hyperparameter Tuning using GridSearch
Step 2: Training
a. Logistic Regressor
b. K-nearest Neighbors
c. Support Vector Classifier
d. Random Forest Classifier
e. Extra Trees Classifier (BEST f1/auc score)
f. Stochastic Gradient Descent Classifier