app.py

import gradio as gr
import numpy as np
import pandas as pd
import pickle
import matplotlib.pyplot as plt

# Load pre-trained models and encoders
with open('is_fraud.pkl', 'rb') as f:
    is_fraud_model = pickle.load(f)

with open('fraud_type.pkl', 'rb') as f:
    fraud_type_model = pickle.load(f)

# Load label encoders for features
encoders = {}
for feature in ['card_type', 'location', 'purchase_category', 'time_of_day']:
    with open(f'{feature}.pkl', 'rb') as f:
        encoders[feature] = pickle.load(f)

with open('fraud_type_le.pkl', 'rb') as f:
    fraud_type_encoder = pickle.load(f)

# Define the prediction function
def predict_fraud(amount, card_type, location, purchase_category, customer_age, time_of_day):
    # Preprocess input
    input_data = pd.DataFrame({
        'amount': [amount],
        'card_type': [card_type],
        'location': [location],
        'purchase_category': [purchase_category],
        'customer_age': [customer_age],
        'time_of_day': [time_of_day]
    })

    # Label encode the inputs
    for feature in ['card_type', 'location', 'purchase_category', 'time_of_day']:
        input_data[feature] = encoders[feature].transform(input_data[feature])

    # Fraud Detection (Binary Classification)
    is_fraud_pred_prob = is_fraud_model.predict_proba(input_data)[0]
    is_fraud_class = np.argmax(is_fraud_pred_prob)

    # Fraud Type Classification (Multiclass Classification)
    fraud_type_pred_prob = fraud_type_model.predict_proba(input_data)[0]
    fraud_type_class = fraud_type_encoder.inverse_transform([np.argmax(fraud_type_pred_prob)])[0]

    # Bar charts for probabilities
    fig, axes = plt.subplots(1, 2, figsize=(12, 5))

    # Plot Fraud Detection Probabilities
    axes[0].bar(['Not Fraudulent', 'Fraudulent'], is_fraud_pred_prob)
    axes[0].set_title('Fraud Detection Probability')
    axes[0].set_ylabel('Probability')
    if is_fraud_class==1:
    # Plot Fraud Type Probabilities
        fraud_types = fraud_type_encoder.classes_
        axes[1].bar(fraud_types, fraud_type_pred_prob)
        axes[1].set_title('Fraud Type Probability')
        axes[1].set_ylabel('Probability')
        plt.xticks(rotation=45, ha="right")

    plt.tight_layout()

    return f"Fraud Detection( {max(is_fraud_pred_prob)*100}% Confidence): {'Fraudulent' if is_fraud_class == 1 else 'Not Fraudulent'}", f"Predicted Fraud Type: {fraud_type_class}" if fraud_type_class==1 else "", fig

# Gradio Interface Setup
interface = gr.Interface(
    fn=predict_fraud,
    inputs=[
        gr.Number(label="Transaction Amount"),
        gr.Dropdown(choices=['MasterCard', 'Visa' ,'Rupay'],label="Card Type"),
        gr.Dropdown(choices=['Surat' ,'Hyderabad' ,'Kolkata' ,'Mumbai' ,'Delhi' ,'Chennai', 'Jaipur',
 'Ahmedabad', 'Bangalore' ,'Pune'],label="Location"),
        gr.Dropdown(choices=['POS' ,'Digital'],label="Purchase Category"),
        gr.Number(label="Customer Age"),
        gr.Dropdown(choices=['night', 'morning', 'afternoon', 'evening'],label="Time of Day")
    ],
    outputs=[
        gr.Markdown(label="Fraud Detection Result"),
        gr.Markdown(label="Fraud Type Result"),
        gr.Plot(label="Probabilities Bar Chart")
    ],
    title="Fraud Detection & Fraud Type Prediction",
    description="By detecting fraudulent transactions, this model helps protect users and businesses in the Indian digital landscape, fostering a safer and more trustworthy online environment.",theme=gr.themes.Soft())

# Launch Gradio Interface
interface.launch()