import warnings from statsmodels.tools.sm_exceptions import ValueWarning warnings.filterwarnings("ignore") warnings.simplefilter(action="ignore", category=UserWarning) warnings.simplefilter(action="ignore", category=FutureWarning) warnings.simplefilter('ignore', ValueWarning) # Interactive plots embedded within the notebook #%matplotlib notebook # Static images of plots embedded within the notebook %matplotlib inline %config InlineBackend.figure_formats = {'png', 'retina'} from IPython.core.interactiveshell import InteractiveShell InteractiveShell.ast_node_interactivity = "all"

import pandas as pd churn_df = pd.read_excel('data/tree-based.xlsx', sheet_name='CHURN')

churn_df

churn_df.info() churn_df.describe(include='all')

df = churn_df.copy()

df.dropna(inplace=True) if 'phone_number' in df: df.drop(columns=['phone_number'], inplace=True) if 'state' in df: df.drop(columns=['state'], inplace=True) if 'area_code' in df: df.drop(columns=['area_code'], inplace=True)

international_plan_cat = ['no', 'yes'] if df['international_plan'].dtype == "object": df['international_plan'] = df['international_plan'].apply(lambda x: international_plan_cat.index(x)) voice_mail_plan_cat = ['no', 'yes'] if df['voice_mail_plan'].dtype == "object": df['voice_mail_plan'] = df['voice_mail_plan'].apply(lambda x: international_plan_cat.index(x)) df['churn'] = df['churn'].astype(int)

X = df.drop('churn', axis=1) Y = df[['churn']]

from sklearn.model_selection import train_test_split X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.3)

from sklearn.tree import DecisionTreeClassifier dt_clf = DecisionTreeClassifier() dt_clf.fit(X_train, y_train)

from sklearn.metrics import f1_score f1_score(y_test, dt_clf.predict(X_test)) from sklearn.metrics import classification_report print(classification_report(y_test, dt_clf.predict(X_test))) from sklearn.metrics import confusion_matrix confusion_matrix(y_test, dt_clf.predict(X_test))

from sklearn import tree text_representation = tree.export_text(dt_clf, feature_names=list(X_train.columns)) print(text_representation)

import graphviz from sklearn import tree dot_data = tree.export_graphviz(dt_clf, out_file=None, feature_names=X.columns, class_names=churn_df['churn'].unique().astype(str), impurity=False, filled=True, rounded=True, proportion=True) # Draw graph graph = graphviz.Source(dot_data, format='png', filename="UntunedModel.gv") graph

feature_list = pd.DataFrame({'feature': X.columns, 'value': dt_clf.feature_importances_}) feature_list_sorted = feature_list.sort_values('value') feature_list_sorted

import matplotlib.pyplot as plt plt.figure(figsize=(7, 3)) plt.barh(range(0, len(feature_list_sorted.index)), feature_list_sorted.value, tick_label=feature_list_sorted.feature) plt.tight_layout();

from sklearn.tree import DecisionTreeClassifier from sklearn.model_selection import GridSearchCV tree_para = { "criterion": ["gini", "entropy", "log_loss"], "max_depth": [5, 8, 9, 10, 11, 12], "max_features": ["auto", "sqrt", "log2"], "splitter": ["random", "best"], } dt_gridsearch = GridSearchCV(DecisionTreeClassifier(), tree_para, scoring="f1") dt_gridsearch.fit(X_train, y_train)

# Sort scores pd.DataFrame(dt_gridsearch.cv_results_).sort_values('mean_test_score', ascending=False) dt_gridsearch.best_params_ # Tuned hyperparameters dt_gridsearch.best_score_ # Best score dt_gridsearch.best_estimator_ # Tuned model

dt_clf_tuned = dt_gridsearch.best_estimator_

from sklearn.tree import DecisionTreeClassifier from sklearn.model_selection import cross_validate scoring_metrics = ["accuracy", "precision_macro", "recall_macro", "f1_macro", "f1"] score_tuned = cross_validate( dt_clf_tuned, X, Y, scoring=scoring_metrics, cv=5, return_train_score=False, ) scores_tuned_df = pd.DataFrame(score_tuned) scores_tuned_df scores_tuned_df.mean()

scores = cross_validate( dt_clf, X, Y, scoring=scoring_metrics, cv=5, return_train_score=False, ) scores_df = pd.DataFrame(scores) scores_df scores_df.mean()

dot_data_tuned = tree.export_graphviz(dt_clf_tuned, out_file=None, feature_names=X.columns, class_names=churn_df['churn'].unique().astype(str), impurity=False, filled=True, rounded=True, proportion=True) # Draw graph graph_tuned = graphviz.Source(dot_data_tuned, format='png', filename="TunedModel.gv") graph_tuned