ml model

.gitignore

@@ -0,0 +1,5 @@
+__pycache__
+build
+barclays_credit_classifier.egg-info
+.env
+

main.py

@@ -0,0 +1,77 @@
+import models
+import pandas as pd
+import preprocessing as pp
+from sklearn.metrics import accuracy_score
+
+# Load Data
+credit_risk = pd.read_csv("credit_risk_dataset.csv")
+
+# Feature Addition
+
+
+# Feature Conversion
+
+person_home_ownership_values = {
+    "RENT": 1,
+    "MORTGAGE": 2,
+    "OWN": 3,
+    "OTHER": 4,
+}
+loan_intent_values = {
+    "EDUCATIONAL": 1,
+    "MEDICAL": 2,
+    "VENTURE": 3,
+    "PERSONAL": 4,
+    "DEBTCONSOLIDATION": 5
+}
+loan_grade_values = {
+    "A": 1,
+    "B": 2,
+    "C": 3,
+    "D": 4,
+    "E": 5
+}
+cb_person_default_on_file_values = {
+    "Y": 1,
+    "N": 0,
+}
+
+# Convert categorical column to a numerical column
+credit_risk["person_home_ownership"] = credit_risk["person_home_ownership"].map(person_home_ownership_values)
+credit_risk["loan_intent"] = credit_risk["loan_intent"].map(loan_intent_values)
+credit_risk["loan_grade"] = credit_risk["loan_grade"].map(loan_grade_values)
+credit_risk["cb_person_default_on_file"] = credit_risk["cb_person_default_on_file"].map(cb_person_default_on_file_values)
+
+print("Feature Conversion Complete")
+
+# Feature Removal
+# columns_for_removal = ["housing_median_age", "total_rooms", "total_bedrooms"]
+# for column in columns_for_removal:
+#    housing.drop(column, axis=1, inplace=True)
+
+# Preprocessing
+credit_risk = pp.impute_missing_values(credit_risk)  # Handle missing values
+print("Missing Values handling Complete")
+# housing = pp.remove_outliers(housing) #Remove outliers
+
+# Training and Testing Preperation
+training_features, training_target_value, test_features, test_target_value = pp.training_test_split(credit_risk, "loan_status")  # Split the data into Training and Test sets
+print("Training and Test features split Complete")
+
+# Normalise the data
+training_features, test_features = pp.normalise(training_features,
+                                                test_features)
+print("Normalisation Complete")
+# Init Models
+rf_model = models.random_forest_classifier(training_features,
+                                           training_target_value)
+print("Model Init Complete")
+
+# Get Predictions
+rf_predictions = rf_model.predict(test_features)
+print("Predictions Complete")
+
+# Compare Results
+accuracy = accuracy_score(test_target_value, rf_predictions)
+print(f"Accuracy: {accuracy}")
+print(rf_predictions)

models.py

@@ -0,0 +1,9 @@
+from sklearn.ensemble import RandomForestClassifier
+
+
+def random_forest_classifier(training_features, training_target):
+    model = RandomForestClassifier(max_features="log2",
+                                   random_state=79,
+                                   n_jobs=-1)
+    model.fit(training_features, training_target)
+    return model

preprocessing.py

@@ -0,0 +1,36 @@
+import pandas as pd
+from scipy import stats
+from sklearn.impute import KNNImputer
+from sklearn.model_selection import train_test_split
+from sklearn.preprocessing import MinMaxScaler
+
+
+def training_test_split(data_frame, target_variable):
+    data_train, data_test = train_test_split(data_frame, train_size=0.75,
+                                             random_state=79, shuffle=True)
+    training_target = data_train[target_variable]
+    test_target = data_test[target_variable]
+    training_features = data_train.drop(target_variable, axis=1)
+    test_features = data_test.drop(target_variable, axis=1)
+    return training_features, training_target, test_features, test_target
+
+
+def normalise(training_features, test_features):
+    scaler = MinMaxScaler().fit(training_features)
+    return scaler.transform(training_features), scaler.transform(test_features)
+
+
+def impute_missing_values(data_frame):
+    imputer = KNNImputer(n_neighbors=9, weights="uniform",
+                         metric="nan_euclidean")
+    imputed_data = imputer.fit_transform(data_frame)
+    return pd.DataFrame(imputed_data, columns=data_frame.columns)
+
+
+def remove_outliers(data_frame):
+    for column in data_frame.columns:
+        z_scores = stats.zscore(data_frame[column])
+        threshold = 1.3412
+        data_frame = data_frame[(z_scores < threshold) &
+                                (z_scores > -threshold)]
+    return data_frame

setup.py

@@ -0,0 +1,18 @@
+from setuptools import setup, find_packages
+
+setup(
+        name="barclays_credit_classifier",
+        version="1.0.0",
+        description="Predicts whether someone will default on their loan. Uses the Credit Risk Dataset from Kaggle",
+        author="r0r-5chach", 
+        author_email="r0r-5chach.xyz@proton.me",
+        packages=find_packages(),
+        install_requires=[
+            "numpy",
+            "scipy",
+            "matplotlib",
+            "pandas",
+            "scikit-learn",
+            "seaborn"
+            ]
+        )

visualisations.py

@@ -0,0 +1,66 @@
+import matplotlib.pyplot as plt
+import seaborn as sns
+
+def missing_value_pairwise_plots(data_frame, null_column, save=False):
+    not_missing_data = data_frame.dropna()
+    mising_data = data_frame[data_frame[null_column].isnull()]
+    for column in data_frame.columns:
+        if column != null_column:
+            plt.figure()
+            plt.title(f"Scatter Plot of {column} against {null_column}")
+            plot_missing_values(column)
+            plt.scatter(not_missing_data[column], not_missing_data[null_column], color=[[0.502, 0, 0.502, 0.4]], label="Existing Values")
+            plt.xlabel(column)
+            plt.ylabel(null_column)
+            plt.legend()
+            if save:
+                plt.savefig(f"missing_values[{column}:{null_column}].png")
+            else:
+                plt.show()
+            plt.close()
+
+def plot_missing_values(column):
+    plt.plot([], [], color="red", alpha=0.4, label="Missing Values")
+    for value in column:
+        plt.axvline(x=value, color="red", alpha=0.4)
+
+def correlation_matrix(data_frame, save=False):
+    matrix = data_frame.corr()
+    plt.figure()
+    sns.heatmap(matrix, annot=True)
+    plt.title("Correlation Matrix of Existing Features")
+    
+    if save:
+        plt.savefig("correlation_matrix.png")
+    else:
+        plt.show()
+    plt.close()
+
+def imputation_plots(data_frame, imputed_data, null_column, columns, save=False):
+    not_missing_data = data_frame.dropna()
+    for column in columns:
+        plt.figure()
+        plt.scatter(imputed_data[column], imputed_data[null_column], color=[[0, 0.502, 0, 0.4]], label="Imputed Data")
+        plt.scatter(not_missing_data[column], not_missing_data[null_column], colot=[[0.502, 0, 0.502, 0,4]], label="Original Data")
+        plt.title(f"Scatter Plot of {column} against {null_column} after KNN(9) Imputation")
+        plt.xlabel(column)
+        plt.ylabel(column)
+        plt.legend()
+        if save:
+            plt.savefig(f"imputation_results[{column}:{null_column}].png")
+        else:
+            plt.show()
+        plt.close()
+
+def outlier_box_plots(data_frame, save=False):
+    for column in data_frame.columns:
+        plt.figure()
+        plt.title(f"Box Plot of {column}")
+        plt.boxplot(data_frame[column])
+        plt.ylabel(column)
+        plt.xticks(rotation=45)
+        if save:
+            plt.savefig(f"outlier_box_plot[{column}].png")
+        else:
+            plt.show()
+        plt.close()