app.py

import streamlit as st
import pandas as pd
import numpy as np
import joblib
import shap
import xgboost as xgb

# Load the saved model and preprocessing objects
model_xgb = joblib.load('model_xgb.joblib')
scaler = joblib.load('scaler.joblib')
ohe = joblib.load('ohe.joblib')

# Extract the unique values for 'country' and 'sector' from the OneHotEncoder (ohe) object
unique_sectors = ohe.categories_[0]  

# Define a mapping of encoded values to repayment interval labels
repayment_interval_mapping = {0: '🚅 Bullet Repayment Interval', 1: '🪙 Irregular Repayment Interval', 2: '📅 Monthly Repayment Interval'}

# Title with emojis and colors
st.markdown("<h1 style='text-align: center; color: blue;'>📊 Loan Repayment Interval Prediction for Phillipines(🇵🇭)📈</h1>", unsafe_allow_html=True)

# Input Features Section with emojis and description
st.write("## 🎯 Input Features")
st.markdown("Please select the variables to predict the repayment interval based on historical data. The model assumes that all loans used for training were successfully repaid. Kindly provide the following information:")

# User input fields using unique country and sector values from the OneHotEncoder object
sector = st.selectbox('🏢 Sector', unique_sectors)
funded_amount = st.number_input('💰 Funded Amount', min_value=0, max_value=20000, value=1000, step=50)
lender_count = st.number_input('👥 Lender Count', min_value=1, max_value=200, value=2, step=1)

# Create a sample observation from the user input
sample_listing = pd.DataFrame({
    'sector': [sector],
    'funded_amount': [funded_amount],
    'lender_count': [lender_count],
})

# Separate categorical and numerical features
cat_features = ['sector']
num_features = ['funded_amount', 'lender_count']

# Get the feature names from the OneHotEncoder
ohe_feature_names = ohe.get_feature_names_out(cat_features)

# Combine numerical feature names with encoded categorical feature names
feature_names = np.concatenate([num_features, ohe_feature_names])

# One-hot encode categorical features
X_cat = pd.DataFrame(ohe.transform(sample_listing[cat_features]), columns=ohe_feature_names)

# Scale numerical features
X_num = pd.DataFrame(scaler.transform(sample_listing[num_features]), columns=num_features)

# Combine processed features
X_processed = pd.concat([X_num, X_cat], axis=1)

# Make a prediction (returns the encoded value)
predicted_encoded_repayment_interval = model_xgb.predict(X_processed)[0]

# Map the encoded value back to the actual repayment interval label
predicted_repayment_interval = repayment_interval_mapping.get(int(predicted_encoded_repayment_interval), "Unknown")

# Display the actual repayment interval label with more style
st.title("✅ Predicted Repayment Interval:")
st.markdown(f"<h2 style='color:green;'>{predicted_repayment_interval}</h2>", unsafe_allow_html=True)

# Explanation for SHAP force plots
st.write("## 🔍 SHAP Explanation")
st.markdown("The following SHAP plot explains the model's decision for the predicted repayment interval. This visualization helps you understand the key features that influenced the model's prediction. The red features push the result higher and the blue ones push it lower.")

# SHAP explanations
explainer = shap.TreeExplainer(model_xgb)
shap_values = explainer.shap_values(X_processed)

# Function to add background color to SHAP plots
def add_background(html_content):
    white_background_style = "<style>body { background-color: white; }</style>"
    return white_background_style + html_content

# Generate and display SHAP force plot based on the predicted repayment interval
if predicted_encoded_repayment_interval == 0:
    st.write("### 🚅 SHAP Force Plot for Bullet Repayment Interval")
    shap_html_path_0 = "shap_force_plot_class_0.html"
    shap.save_html(shap_html_path_0, shap.force_plot(
        explainer.expected_value[0],
        shap_values[0][:, 0],
        X_processed.iloc[0, :].values,
        feature_names,
        show=False,
        matplotlib=False
    ))
    with open(shap_html_path_0, 'r', encoding='utf-8') as f:
        shap_html_0 = f.read()
    st.components.v1.html(add_background(shap_html_0), height=130)

elif predicted_encoded_repayment_interval == 1:
    st.write("### 🪙 SHAP Force Plot for Irregular Repayment Interval")
    shap_html_path_1 = "shap_force_plot_class_1.html"
    shap.save_html(shap_html_path_1, shap.force_plot(
        explainer.expected_value[1],
        shap_values[0][:, 1],
        X_processed.iloc[0, :].values,
        feature_names,
        show=False,
        matplotlib=False
    ))
    with open(shap_html_path_1, 'r', encoding='utf-8') as f:
        shap_html_1 = f.read()
    st.components.v1.html(add_background(shap_html_1), height=130)

elif predicted_encoded_repayment_interval == 2:
    st.write("### 📅 SHAP Force Plot for Monthly Repayment Interval")
    shap_html_path_2 = "shap_force_plot_class_2.html"
    shap.save_html(shap_html_path_2, shap.force_plot(
        explainer.expected_value[2],
        shap_values[0][:, 2],
        X_processed.iloc[0, :].values,
        feature_names,
        show=False,
        matplotlib=False
    ))
    with open(shap_html_path_2, 'r', encoding='utf-8') as f:
        shap_html_2 = f.read()
    st.components.v1.html(add_background(shap_html_2), height=130)