Skip to content

ibodumas/gscv-ensemble-voting

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Ensemble Voting (Grid Search CV)

@@FUN STUFF

:bowtie: 💥 🙏 🏃

Screenshot

Pre-processing:

Step 1: Loaded the training dataset into a pandas data frame.
Step 2: Removed redundant features
Step 3: Calculated the percentage of the missing values per features.
Step 4: Eliminated the features with very high ratio of missing values.
Step 5. Constructed imputation processes by most frequent for handling the remaining features with missing values.
Step 6: Built a procedure to properly incorporate the text based feature;
        a) Take out the variable.
        b) Get all the unique codes across all the samples.
        c) Made a data frame with columns corresponding to (b) above.
        d) Went through the samples, split the codes, and flagged the columns corresponding to the codes in (c) above.
        e) These resulted in a data frame with the columns as unique codes. And there values are 1 or 0 signifying the presence of the code or not respectively.
Step 7: Concatenated the processed feature data with the remaining data set.
Step 8: Initiated One-Hot encoding for the categorical variables (ignored ordering).
Step 9: Implemented [0, 1] normalization
Step 10: Initiated PCA with about 99% variance explained, however it didn’t perform well on the models. Hence, it was dropped.

Modeling:

Step 1: Split the preprocessed training data into train set and test set.
Step 2: Picked 10 for k-fold cross validation (This can be an environment variable).
Step 3: Set metric evaluation to roc-auc (This can be an environment variable).
alt text

Step 4: Constructed a data frame to house the performance of each models.
Step 5: Implemented grid search cross validation for the models that requires picking optimal hyper-parameters.
Step 5: Models
        a) Naive Bayes
        b) KNN
        c) Logistic Regression – Ridge
        d) SVM - RBF Kernel
        e) Multi-layer Perceptron (MLP)
Step 6: Built an ensemble method by voting for combining all the models.