Script runs SVN and Desicion Tree models to predict customers churn.

EDA, vizualization and different models exploration are in exploratory_data_analysis.ipynb

Short summary of analysis:

                    Model       Score       ROC

0 Support Vector Machines 0.987241 0.987272 1 KNN 92.980000 0.859904 3 Random Forest 100.000000 0.813813 4 Naive Bayes 80.410000 0.802328 9 Decision Tree unprunned 100.000000 0.771798 8 Decision Tree 92.430000 0.770063 2 Logistic Regression 76.280000 0.762366 5 Perceptron 59.360000 0.588731 7 Linear SVC 51.920000 0.519723 6 Stochastic Gradient Decent 49.990000 0.498934

Feature importance:

           importance

DayMins 0.152191 CustServCalls 0.126595 DayCharge 0.125896 EveCharge 0.069660 EveMins 0.065197 IntlPlan 0.061667 IntlCalls 0.049894 IntlMins 0.045316 IntlCharge 0.034040 NightMins 0.033756 DayCalls 0.031391 NightCharge 0.031391 AccountLength 0.030978 EveCalls 0.030150 NightCalls 0.028654 VMailMessage 0.027222 State 0.024262 VMailPlan 0.024155 AreaCode 0.007586

Also in repositary are US State churn interactive vizualization and pdf for desesion tree model.