edit_notes.md

In the original project three different classification methods have been used. Building upon that, four more methods have been implemented with improved in performance. Further, other metrics like f1 score and jaccard score have also implemented for analysis of performance.

See edit_main.py for changes.

In the edit, seven different methods have been using for classification:

• Decision Tree Classification
• Random Forest Classification
• Gradient Boosting Classification
• Extra Trees Classification
• Bagging Classification
• Adaptive Boost Classification with a DecisionTreeClassifier as the base estimator
• Adaptive Boost Classification with a RandomForestClassifier as the base estimator

The results have been presented:

Training Data	Decision Tree	Random Forest	Gradient Boost	Extra Trees	Bagging Classifier	Adaptive Boost D	Adaptive Boost R
Accuracy	0.871425127	0.87012298	0.877525922	0.851977333	0.998794309	0.883349409	0.887834579
F1 Score	0.928212723	0.927430004	0.931319387	0.918715282	0.999281578	0.93379047	0.936760317
Jaccard Score	0.866041931	0.864680167	0.871466532	0.849651591	0.998564188	0.875803905	0.881043412
Test Data	Decision Tree	Random Forest	Gradient Boost	Extra Trees	Bagging Classifier	Adaptive Boost D	Adaptive Boost R
Accuracy	0.86727755	0.866409453	0.870894623	0.851073065	0.907595852	0.870653484	0.877212443
F1 Score	0.925949844	0.925397253	0.927701407	0.918168327	0.946910502	0.926737325	0.93097278
Jaccard Score	0.862110432	0.861152882	0.865152126	0.848716441	0.899173815	0.863476712	0.870859751

Note: Adaptive Boost D stands for Adaptive Boost Classification with a DecisionTreeClassifier as the base estimator
Adaptive Boost R stands for Adaptive Boost Classification with a RandomForestClassifier as the base estimator

Immediate Conclusions:

Bagging Classifier seems to work the best with a 0.907595852423438 accuracy over the testing data and a 0.998794309139136 accuracy over the training data

Note: See Customer Seg Performance Analysis.xlsx for results