This experiment performs predictive analysis on Agency Attrition data to decide whether agent attritions or remains in force.
Shape = 1000 x 188. Out of the 187 columns, 1 for label, 31 features are manually chosen based on predetermined importance decided by their ranks (1, 2). Next, binary classifiers are applied to the data for training. The train and test rows are chosen using 10 fold cross validation and split data for different classifiers. The confusion matrix is observed to give a very few false positives and false negatives thereby taking the AUC to almost 0.97.