Upload 5 files

app.py

@@ -0,0 +1,181 @@
+"""
+Loan Default Prediction — Streamlit App
+Deployed on Hugging Face Spaces
+"""
+
+import streamlit as st
+import pandas as pd
+import numpy as np
+import joblib
+import json
+import shap
+import matplotlib.pyplot as plt
+
+# ─── Page Config ───
+st.set_page_config(
+    page_title="Loan Default Prediction",
+    page_icon="🏦",
+    layout="wide"
+)
+
+# ─── Load Artifacts ───
+@st.cache_resource
+def load_artifacts():
+    model = joblib.load("xgb_model.joblib")
+    scaler = joblib.load("scaler.joblib")
+    with open("selected_features.json") as f:
+        features = json.load(f)
+    background = joblib.load("shap_background.joblib")
+    explainer = shap.TreeExplainer(model)
+    return model, scaler, features, background, explainer
+
+model, scaler, feature_names, background, explainer = load_artifacts()
+
+# ─── Title ───
+st.title("🏦 Loan Default Prediction")
+st.markdown("""
+This app predicts whether a loan applicant is likely to **default** or be **approved**,
+using an XGBoost model trained on 45,000 loan records. It also provides a SHAP-based
+explanation of each prediction.
+""")
+
+st.divider()
+
+# ─── Sidebar Inputs ───
+st.sidebar.header("📝 Applicant Information")
+
+person_age = st.sidebar.slider("Age", 18, 80, 30)
+person_income = st.sidebar.number_input("Annual Income ($)", 8000, 500000, 50000, step=1000)
+person_emp_exp = st.sidebar.slider("Employment Experience (years)", 0, 60, 5)
+loan_amnt = st.sidebar.number_input("Loan Amount ($)", 500, 100000, 10000, step=500)
+loan_int_rate = st.sidebar.slider("Loan Interest Rate (%)", 2.0, 25.0, 10.0, step=0.1)
+loan_percent_income = st.sidebar.slider("Loan as % of Income", 0.0, 1.0, 0.2, step=0.01)
+cb_person_cred_hist_length = st.sidebar.slider("Credit History Length (years)", 1.0, 30.0, 5.0, step=0.5)
+credit_score = st.sidebar.slider("Credit Score", 300, 850, 650)
+
+previous_loan_defaults = st.sidebar.selectbox("Previous Loan Defaults?", ["No", "Yes"])
+person_education = st.sidebar.selectbox("Education Level", ["High School", "Associate", "Bachelor", "Master", "Doctorate"])
+person_home_ownership = st.sidebar.selectbox("Home Ownership", ["RENT", "OWN", "MORTGAGE", "OTHER"])
+loan_intent = st.sidebar.selectbox("Loan Intent", ["PERSONAL", "EDUCATION", "MEDICAL", "VENTURE", "HOMEIMPROVEMENT", "DEBTCONSOLIDATION"])
+
+# ─── Feature Engineering ───
+debt_to_income_ratio = loan_amnt / person_income if person_income > 0 else 0
+
+# Age group
+if person_age <= 25:
+    age_group = "Young"
+elif person_age <= 35:
+    age_group = "Adult"
+elif person_age <= 50:
+    age_group = "Middle_Age"
+else:
+    age_group = "Senior"
+
+# Income category
+if person_income <= 30000:
+    income_cat = "Low"
+elif person_income <= 60000:
+    income_cat = "Medium"
+elif person_income <= 100000:
+    income_cat = "High"
+else:
+    income_cat = "Very_High"
+
+# ─── Build Feature Vector ───
+# Must match exact feature order from training
+input_dict = {
+    'person_age': person_age,
+    'person_income': person_income,
+    'person_emp_exp': person_emp_exp,
+    'loan_amnt': loan_amnt,
+    'loan_int_rate': loan_int_rate,
+    'loan_percent_income': loan_percent_income,
+    'cb_person_cred_hist_length': cb_person_cred_hist_length,
+    'credit_score': credit_score,
+    'previous_loan_defaults_on_file': 1 if previous_loan_defaults == "Yes" else 0,
+    'debt_to_income_ratio': debt_to_income_ratio,
+    # One-hot: person_education (drop_first = Associate)
+    'person_education_Bachelor': 1 if person_education == "Bachelor" else 0,
+    'person_education_Doctorate': 1 if person_education == "Doctorate" else 0,
+    'person_education_High School': 1 if person_education == "High School" else 0,
+    'person_education_Master': 1 if person_education == "Master" else 0,
+    # One-hot: home_ownership (drop_first = MORTGAGE)
+    'person_home_ownership_OTHER': 1 if person_home_ownership == "OTHER" else 0,
+    'person_home_ownership_OWN': 1 if person_home_ownership == "OWN" else 0,
+    'person_home_ownership_RENT': 1 if person_home_ownership == "RENT" else 0,
+    # One-hot: loan_intent (drop_first = DEBTCONSOLIDATION)
+    'loan_intent_EDUCATION': 1 if loan_intent == "EDUCATION" else 0,
+    'loan_intent_HOMEIMPROVEMENT': 1 if loan_intent == "HOMEIMPROVEMENT" else 0,
+    'loan_intent_MEDICAL': 1 if loan_intent == "MEDICAL" else 0,
+    'loan_intent_PERSONAL': 1 if loan_intent == "PERSONAL" else 0,
+    'loan_intent_VENTURE': 1 if loan_intent == "VENTURE" else 0,
+    # One-hot: age_group (drop_first = Young)
+    'age_group_Adult': 1 if age_group == "Adult" else 0,
+    'age_group_Middle_Age': 1 if age_group == "Middle_Age" else 0,
+    'age_group_Senior': 1 if age_group == "Senior" else 0,
+    # One-hot: income_category (drop_first = Low)
+    'income_category_Medium': 1 if income_cat == "Medium" else 0,
+    'income_category_High': 1 if income_cat == "High" else 0,
+    'income_category_Very_High': 1 if income_cat == "Very_High" else 0,
+}
+
+input_df = pd.DataFrame([input_dict])[feature_names]
+
+# Scale
+input_scaled = pd.DataFrame(scaler.transform(input_df), columns=feature_names)
+
+# ─── Predict ───
+if st.sidebar.button("🔍 Predict", type="primary", use_container_width=True):
+    prediction = model.predict(input_scaled)[0]
+    probability = model.predict_proba(input_scaled)[0]
+
+    col1, col2 = st.columns(2)
+
+    with col1:
+        st.subheader("Prediction Result")
+        if prediction == 1:
+            st.error(f"⚠️ **LOAN DEFAULT** — Probability: {probability[1]*100:.1f}%")
+        else:
+            st.success(f"✅ **LOAN APPROVED** — Probability: {probability[0]*100:.1f}%")
+
+        st.metric("Default Probability", f"{probability[1]*100:.1f}%")
+        st.metric("Approval Probability", f"{probability[0]*100:.1f}%")
+
+    with col2:
+        st.subheader("SHAP Explanation")
+        shap_values = explainer.shap_values(input_scaled)
+
+        shap_explanation = shap.Explanation(
+            values=shap_values[0],
+            base_values=explainer.expected_value,
+            data=input_scaled.iloc[0].values,
+            feature_names=feature_names
+        )
+
+        fig, ax = plt.subplots(figsize=(8, 6))
+        shap.plots.waterfall(shap_explanation, show=False)
+        st.pyplot(fig)
+        plt.close()
+
+    # Feature contributions table
+    st.subheader("Feature Contributions")
+    contrib_df = pd.DataFrame({
+        'Feature': feature_names,
+        'Input Value': input_df.iloc[0].values,
+        'SHAP Value': shap_values[0]
+    }).sort_values('SHAP Value', key=abs, ascending=False)
+    contrib_df['Direction'] = contrib_df['SHAP Value'].apply(
+        lambda x: '↑ Increases Default Risk' if x > 0 else '↓ Decreases Default Risk'
+    )
+    st.dataframe(contrib_df, use_container_width=True, hide_index=True)
+
+else:
+    st.info("👈 Fill in the applicant details in the sidebar and click **Predict**.")
+
+# ─── Footer ───
+st.divider()
+st.markdown("""
+**Model:** XGBoost (200 estimators, max_depth=6) | **Accuracy:** 92.78% | **ROC-AUC:** 0.9757
+
+**Explainability:** SHAP (TreeExplainer) for post-hoc explanations of the black-box model.
+""")

scaler.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f58c0d8c36dbcab46a42b4b9ea358d4af3a16672e18e37fb9bc2ac6b85c370ee
+size 2199

selected_features.json

@@ -0,0 +1,30 @@
+[
+  "person_age",
+  "person_income",
+  "person_emp_exp",
+  "loan_amnt",
+  "loan_int_rate",
+  "loan_percent_income",
+  "cb_person_cred_hist_length",
+  "credit_score",
+  "previous_loan_defaults_on_file",
+  "debt_to_income_ratio",
+  "person_education_Bachelor",
+  "person_education_Doctorate",
+  "person_education_High School",
+  "person_education_Master",
+  "person_home_ownership_OTHER",
+  "person_home_ownership_OWN",
+  "person_home_ownership_RENT",
+  "loan_intent_EDUCATION",
+  "loan_intent_HOMEIMPROVEMENT",
+  "loan_intent_MEDICAL",
+  "loan_intent_PERSONAL",
+  "loan_intent_VENTURE",
+  "age_group_Adult",
+  "age_group_Middle_Age",
+  "age_group_Senior",
+  "income_category_Medium",
+  "income_category_High",
+  "income_category_Very_High"
+]

shap_background.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5394e9082e06f2e94d26041a1399aed5ad9fe12b88a5abdd11f7350911d077e
+size 26923

xgb_model.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ed6d104b4f07870431e60012354a7218b6e694d2522dc475c71c0fe1fbe7c37
+size 597784