app.py

# app.py
import streamlit as st
import pickle
import pandas as pd
import joblib

# Load model and encoder
try:
    with open('model/Grad_Boost.sav', 'rb') as model_file:
        loaded_model = pickle.load(model_file)   # trained GradientBoost model
    with open('PreProcessing/encoders.pkl', 'rb') as encoder_file:
        loaded_encoder = joblib.load(encoder_file)   # trained TargetEncoder
    st.success("✅ Model and encoder loaded successfully!")
except Exception as e:
    st.error(f"Error loading model/encoder: {e}")

st.title("🚨 Claim Fraud Detection")

# Define input features (MUST match training)
features = [
    'months_as_customer','policy_state','policy_csl','policy_deductable',
    'policy_annual_premium','umbrella_limit','insured_zip','insured_sex',
    'insured_education_level','insured_occupation','insured_hobbies',
    'insured_relationship','capital-gains','capital-loss','incident_type',
    'collision_type','incident_severity','authorities_contacted',
    'incident_state','incident_city','incident_hour_of_the_day',
    'number_of_vehicles_involved','property_damage','bodily_injuries',
    'witnesses','police_report_available','total_claim_amount',
    'injury_claim','property_claim','auto_make','auto_year',
    'incident_month','incident_day','auto_model'
]

st.sidebar.header("Input Features")

# Collect user inputs
input_data = {}
for feature in features:
    if feature in [
        'policy_state','policy_csl','insured_sex','insured_education_level',
        'insured_occupation','insured_hobbies','insured_relationship',
        'incident_type','collision_type','incident_severity','authorities_contacted',
        'incident_state','incident_city','property_damage','police_report_available',
        'auto_make','auto_model'
    ]:
        input_data[feature] = st.sidebar.text_input(f"{feature}")
    else:
        input_data[feature] = st.sidebar.number_input(f"{feature}", value=0.0)

# Predict button
if st.sidebar.button("Predict Fraud"):
    # Convert to DataFrame
    input_df = pd.DataFrame([input_data])

    # Apply trained encoder ONLY on categorical cols
    categorical_cols = [
        'auto_model','policy_state','incident_type','auto_make','police_report_available',
        'property_damage','incident_city','incident_state','authorities_contacted',
        'incident_severity','collision_type','insured_relationship','insured_hobbies',
        'insured_occupation','insured_education_level','insured_sex','policy_csl'
    ]
   
    
    # Encode categorical features
    for col, te in loaded_encoder.items():
        input_df[col] = te.transform(input_df[[col]])

    # Align columns with model training
    expected_cols = loaded_model.feature_names_in_
    input_df = input_df.reindex(columns=expected_cols, fill_value=0)


    # Predict
    prediction = loaded_model.predict(input_df)[0]
    prediction_proba = loaded_model.predict_proba(input_df)[0][1]
   

    st.subheader("🔎 Prediction Result")
    st.write("**Fraudulent Claim** 🚩" if prediction == 1 else "✅ Not Fraudulent")
    st.write(f"Fraud Probability: {prediction_proba:.2f}")