423923, 12538 AM In [1]: amport pandas as pd Loan Prediction - Jupyter Notebook import matplotlib.pyplot as plt import seaborn as sns import nunpy as ap sns.set_theme(color_codes=True) In [2]: df = pd.read_csv('loan_train.csv') Gender Married Dependents Education Self_Employed Applicant_income Coapplicant_Income Loan df.head() out [2]: 0 Male No 1 Male Yes 2 Male Yes 3 Male Yes 4 Male No 0 1 ° Graduate Graduate Graduate Not Graduate Graduate No No Yes No. No Data Preprocessing Part 1 In [3]: #Check the number of unique value on object datatype df. select_dtypes (includ out[3]: Gender Married Dependents Education Self_Enployed Area Status dtype: intea object") nunique() Exploratory Data Analysis localhost 8890Inotebooks/Loan Prediction ayn ‘584900 458300 300000 258300 600000 00 150800.0, 09 236800. 00 1126423923, 12538 AM In [4]: Loan Prediction -Jupyter Notebook # List of categorical variables to plot cat_vars = [‘Gender’, ‘Married’, ‘Dependents’, ‘Education’, ‘self Employed’, ‘Area’, ‘Credit History’, ‘Dependents’] # create figure with subplots Fig, axs = plt.subplots(nrows=2, ncols=4, figsize=(15, 10)) axs = axs.flatten() # create barplot for each categorical variable for i, var in enumerate(cat_vars): sns.countplot(x=var, hue='Status’, datasdf, ax=axs[i]) axs[i].set_xticklabels(axs[i].get_xticklabels(), rotation=90) # adjust spacing between subplots ig. tight_layout() show plot pit. show() z ie C Q i ar localhost 8890Inotebooks/Loan Prediction ayn4123923, 12338 AM In [5]: locahhost 8890inotebooks/Loan Prediction ayn Loan Prediction -Jupyter Notebook import warnings warnings. filterwarnings(" ignore") # get List of categorical variables cat_vars = [‘Gender', ‘Married’, ‘Dependents’, ‘Education’, ‘self_Employed’, ‘Area’, ‘Credit History’, ‘Dependents'] # create figure with subplots fig, axs = plt.subplots(nrows=2, ncols=4, figsize=(15, 10)) axs = axs.flatten() # create histplot for each categorical variable for i, var in enumerate(cat_vars): sns.histplot(x-var, hue='Status', data-df, ax-axs[i], multiple="Fill” axs[i].set_xticklabels(df[var] .unique(), rotation=90) axs[i].set_xlabel (var) kde-False, ‘# adjust spacing between subplots fig. tight_layout() # show plot plt.show() ik | i =. I fi a423923, 12538 AM Loan Prediction -Jupyter Notebook In [6]: num_vars = ['Applicant_Income’, ‘Coapplicant_Income', ‘Term'] Fig, axs = plt.subplots(nrows=1, ncols=3, figsize=(15, 7)) axs = axs.flatten() for i, var in enumerate(num_vars): sns.boxplot(x=var, dat: fig.tight_layout() plt.show() localhost 8890Inotebooks/Loan Prediction ayn 426423923, 12538 AM Loan Prediction -Jupyter Notebook In [7]: num_vars = ['Applicant_Income’, ‘Coapplicant_Income', ‘Term'] Fig, axs = plt-subplots(nrows=1, ncols=3, figsize=(15, 7)) axs = axs.flatten() for i, var in enumerate(num_vars): sns.violinplot(x=var, data=df, a) fig. tight_layout() plt.show() localhost 8890Inotebooks/Loan Prediction ayn423923, 12538 AM Loan Prediction -Jupyter Notebook In [8]: num_vars fig, axs = plt-subplots(nrows=1, ncols=3, figsiz axs = axs.flatten() as, 7) for i, var in enunerate(num_vars): sns.violinplot (x=var, y='Status', data-df, ax-axs[i]) fig. tight_layout() plt.show() iio aa localhost 8890Inotebooks/Loan Prediction ayn ['Applicant_tncome", "Coapplicant_Incone', ‘Term’ ] ) + + 626423923, 12538 AM Loan Prediction -Jupyter Notebook In [9]: num_vars = ['Applicant_Income’, ‘Coapplicant_Income', ‘Term'] Fig, axs = plt-subplots(nrows=1, ncols=3, figsize=(15, 7)) axs = axs.flatten() for i, var in enumerate(num_vars): sns.histplot(x=var, data=df, ai fig.tight_layout() plt.show() 7 lu 7 = . — localhost 8890Inotebooks/Loan Prediction ayn 1128423923, 12538 AM In [10]: num_vars fig, axs Loan Prediction -Jupyter Notebook [‘Applicant_Income’, ‘Coapplicant_Incone’, ‘Term’ ] plt. subplots (nrow: axs.flatten() for i, var in enunerate(num_vars): fig. tight_layout() plt.show() = sns.histplot (: Data Preprocessing Part 2 In [11]: df-head() » ncols=3, figsize=(15, 7)) ar, data=df, hue='Status', ax=axs[i]) oss): Gender Married Dependents Education Self_Employed Applicant Income Coapplicant_Income Loan, 0 Wale We 0 Grote We seeac0 00 1 Mae Yes 1 Grote we asss00 ss00000 2 Mle ves 0 Graduate ves sooneo 00 3 Mle ves © Gat W 2sss00 zsst000 4 tle Wo 0 Grodute " sooito 00 localhost 8890Inotebooks/Loan Prediction ayn 8126423923, 12538 AM In [12]: out [12]: In [13]: out (13): In [14]: out (14): In [15]: Loan Prediction - Jupyter Notebook heck the missing value check missing = df.isnul1().sum() * 100 / df.shape[o] check_missing[check_missing > 0].sort_values(ascending-False) Credit History 8.143322 Self_Employed 5.211726 Dependents 2.442997 Term 2.280130 Gender 2.117264 Married 0.488599 dtype: floatea # FILL null values with ‘Unknown* df.fillna( ‘Unknown’, inplace=True) #check the missing value again check_missing = df.isnull().sum() * 100 / df.shape[2] check_missing|check_missing > @].sort_values(ascending=False) Series({], dtyp float6a) df.dtypes Gender object Married object Dependents object Education object Self _Enployed object Applicant_Incone intes Coapplicant_Income floate4 Loan_Anount intes Term object Credit History object Area object Status object dtype: object Label Encoding for Object datatype # Loop over each column in the DataFrane where dtype is ‘object’ for col in df.select_dtypes(include=[ ‘object’ ]).columns: # Print the column name and the unique values print(#"{col}: {df[col] .unique()}") Gender: ['Male' 'Female' ‘Unknown" ] Married: ['No' "Yes" 'Unknown'] Dependents: ['@" "1" '2' "3+" "Unknown" } Education: ['Graduate’ 'Not Graduate’ ] Self_Employed: ['No’ "Yes" ‘Unknown’ ] Term: [360.0 120.0 240.2 ‘Unknown’ 180.0 60.0 300.0 480.0 36.0 84.0 12.0] Credit History: [1.@ @.@ "Unknown" ] Area: ['Urban' ‘Rural’ ‘Semiurban" ] Status: ['¥" 'N'] localhost 8890Inotebooks/Loan Prediction ayn 926423923, 12538 AM In [16]: In [17]: Loan Prediction - Jupyter Notebook # Convert selected columns to string data type AF[[ Term’, ‘Credit History']] = df[[ ‘Term’, ‘Credit History’ ]].astype(str) from sklearn import preprocessing # Loop over each column in the DataFrame where dtype is ‘object’ for col in df.select_dtypes(include=[ ‘object’ ]).columns: # Initialize a LabelEncoder object label_encoder = preprocessing. LabelEncoder() 4 Fit the encoder to the unique values in the column label_encoder .fit (dF [col] .unique()) 4 Transform the column using the encoder df [col] = label_encoder.transform(df[col]) # Print the column name and the unique encoded values print(#"{col}: {df[col] .unique()}") Gender: [1 @ 2] Married: [@ 2 1] Dependents: [@ 12 3 4] Education: [6 1] Self_Employed: [0 2 1] Terms [6 1310 28475 9 @] Credit History: [1 @ 2] Area: [2 @ 1] Status: [1 0] Check if the Label 'Status' is balanced or not localhost 8890Inotebooks/Loan Prediction ayn 10126423923, 12538 AM Loan Prediction -Jupyter Notebook In [18]: sns.countplot (df[ status" ]) df[ ‘Status’ ].value_counts() out{1s}: 1 422 @ 192 Name: Status, dtype: intes Status Oversampling Minority Class to balance the Label In [19]: from sklearn.utils import resample create tno different dataframe of majority and minority class df_majority = df{(df[ ‘status’ df_minority = df{(dF[ ‘Status’ # upsample minority class df_minority_upsampled = resample(df_minority, replace=true, # sample with replacement ALsamples= 422, # to match majority class randon_state=0) # reproducible results # Combine majority class with upsampled minority class df_upsanpled = pd.concat([d#_minority_upsampled, df majority]) localhost 8890Inotebooks/Loan Prediction ayn 26423923, 12538 AM Loan Prediction -Jupyter Notebook In [20]: sns.countplot (4f_upsampled| ‘status’ ]) df_upsanpled|[ ‘status’ ].value_counts() out[2e}: @ 422 1 422 Name: Status, dtype: intes 400 350 ‘count 150 100 Status Remove Outi r using IQR because there are alot of extreme value In [21]: df_upsampled. shape out(21]: (a4, 12) localhost 8890Inotebooks/Loan Prediction ayn 1226423923, 12538 AM In [22]: # specify the columns to remove outliers from dataframe colunn_names = ['Applicant_Incone’ , “coapplicant_Incone’ Loan Prediction - Jupyter Notebook "Term # remove outliers for each selected column using the IQR method for colunn_nane in column_nanes: d¥_upsampled[ colurn_nane] .quantile(@.25) df_upsanpled[ column_nane] .quantile(9.75) a= @ IQR df_upsanpled.head() G-a 4f_upsanpled = df_upsanpled[~((df_upsampled[colunn_name] < (Qt - 1.5 * TOR) | ( Married Dependents Education Self_Employed Applicant_Incomo Coapplicant_Income Lo out [22]: Gender 148 0 338 ° os 1 57 1 107 1 0 o 2 0 a 1 0 1 ° In [23]: #Check the shape after outlier removal. df_upsanpled. shape out[23]: (614, 12) localhost 8890Inotebooks/Loan Prediction ayn 0 ° 1 ° 1000000 182000 371700 336600 1733300 +166600.0 o. 292500.0 220000.0 °. 1926423923, 12538 AM Loan Prediction -Jupyter Notebook In [24]: plt.figure(Figsize=(15,12)) sns.heatmap(df_upsampled.corr(), fmt: annot=True) out [24]: In [25]: df_upsampled.drop(columns='Term’, inplace=True) Train Test Split 4f_upsampled.drop( ‘Status’, axis=1) ROC Curve 10 08 & True Positive Rate g 02 00 a0 02 a4 06 a8 1.0 False Positive Rate Random Forest 20128423923, 12538 AM In [39]: In [40]: out [40]: In [41]: In [42]: Loan Prediction - Jupyter Notebook from sklearn.ensemble import RandonForestClassifier from sklearn.model_selection import GridSearchcV rfc = RandonForestClassifier() param_grid = { ‘nestimators': [100, 200], *max_depth': [None, 5, 10], ‘max_features': ["sqrt’, ‘log2', None] } # Perform a grid search with cross-validation to find the best hyperparaneters grid_search = GridSearchCv(rfc, paran_grid, cv=5) grid_search.fit(x_train, y_train) # Print the best hyperparameters print (grid_search.best_parans_) {’max_depth': None, 'max_features': 'log2', 'n_estimators': 200) from sklearn.ensenble import RandonForestClassifier rfc = RandomForestClassifier(random_state=0, max_features="log2", n_estimators=200) rfc. fit(X_train, y_train) RandonForestClassifier(max_feature: 1og2", n_estinators=200, random_stat« y_pred = rfc.predict(X_test) print("Accuracy Score :", round(accuracy_score(y_test, y_pred)*100 ,2), "%") Accuracy Score : 95.12 % from sklearn.metrics inport accuracy_score, f1_score, precision_score, recall_score, print(‘F-1 Score : ',(f1_score(y test, y_pred, average='micro'))) print('Precision Score : ',(precision score(y test, y_pred, average='micro’))) print('Recall Score : ',(recall_score(y test, y_pred, average='micro'))) print(‘Jaccard Score : ',(Jaccard_score(y_test, y_pred, average="micro'))) print(*Log Loss : ', (log loss(y_test, y_pred))) F-1 Score : 0.9512195121951219 Precision Score : @.9512195121951219 Recall Score : @.9512195121951219 Jaccard Score : @.9069767441860465 Log Loss : 1.6848443638958128, localhost 8890Inotebooks/Loan Prediction ayn 21128423923, 12538 AM In [43]: localhost 8890Inotebooks/Loan Prediction ayn Loan Prediction -Jupyter Notebook imp, pd.DataFrame({ ‘eature Name": X_train. columns, "Importance": rfc. feature_importances_ » Fi = imp_df.sort_values(by="Inportance", ascending=False) fi2 = fi-head(10) plt. figure(figsize=(10,8)) sns. barplot (dat: ‘Importance’, y='Feature Name’) plt.title('Top 10 Feature Importance Each Attributes (Random Forest)’, fontsize=18) plt.xlabel ("Inportance’, fontsize=16) plt.ylabel (‘Feature Name’, fontsize=16) plt. show() Top 10 Feature Importance Each Attributes (Random Forest) Peplicant income ret History Loon_Amount (coappticant income Dependents ea Feature Name Sell Employed Eveaton 8 010 ons Importance 2226423923, 12538 AM Loan Prediction -Jupyter Notebook In [44]: import shap # compute SHAP values explainer = shap.TreeExplainer(rfc) shap_values = explainer.shap_values(X_test) shap.sunmary plot(shap_values[1], X_test.values, feature_names = X_test.colunns) High Credit History + smeeme e+ Applicant_Income . Loan_Amount tat . Area teehee ° Coapplicant_Income = batterie = 3 Dependents . . 2 Gender ste . Married eh: Self_Employed cote = Education “ 1 1 Pow 46 45 04 03 42 01 00 of 02 SHAP value (impact on model output) localhost 8890Inotebooks/Loan Prediction ayn 2226423923, 12538 AM Loan Prediction -Jupyter Notebook In [45]: import shap explainer = shap.TreeExplainer(rfc) shap_values = explainer.shap_values(X_test) shap.sunmary_plot(shap_values, X_test) Applicant_Income Loan_Amount Area Coapplicant_income Dependents Gender Married Self_Employed | mmm Class 1 mms Class 0 Education 000 «0050005280 mean(/SHAP value|) (average impact on model output magnitude) localhost 8890Inotebooks/Loan Prediction ayn 241264123723, 12:38 AM Loan Presi -Jpyter Notebook In [46]: from sklearn.metrics import confusion matrix cm = confusion matrix(y test, y_pred) plt.figure(figsize=(5,5)) sns-heatmap (data=cm, Linewidths= pit.ylabel(‘Actual label’) plt.xlabel( ‘Predicted label’) all_sanple_title = ‘Accuracy Score for Randon Forest: {@}'.format(rfe.score(X test, y, plt.title(all_sample title, size = 15) 5, annot-True, cmap = ‘Blues') Out[46]: Text(@.5, 1.8, ‘Accuracy Score for Random Forest: @.9512195121951219") Accuracy Score for Random Forest: 0.9512195121951219 60 ° a 4 = 40 30 -20 Actual labo! -10 Predicted label localhost 8890Inotebooks/Loan Prediction ayn 25126423923, 12538 AM In [47]: out (47): localhost 8890Inotebooks/Loan Prediction ayn Loan Prediction - Jupyter Notebook from sklearn.metrics import roc_curve, roc_auc_score y_pred_proba = rfc.predict_proba(X_test)[:][:, df_actual_predicted = pd.concat([pd.DataFrame(np.array(y_test), colunns=["y_actual']) df_actual_predicted.index = y_test. index for, tpr, tr = roc_curve(df_actual_predicted['y_actual'], df_actual_predictedt 'y_pred auc’ = ro¢_auc_score(df_actual_predicted["y actual], df actual_predicted[ "y pred prob, plt.plot(fpr, tpr, label='AUC = %0.4F" Xauc) pit.plot(#pr, fpr, linestyle = '--', color plt.xlabel(‘False Positive Rate’) plt.ylabel(‘True Positive Rate’) plt.title('ROC Curve’, size = 15) plt.legend() kK) ROC Curve 10 08 & True Positive Rate g 02 oo — Avc=09871 or) 02 a4 06 a8 1.0 False Positive Rate 26128