app.py

import gradio as gr
import numpy as np
import pandas as pd
import joblib
import os
import warnings
from huggingface_hub import hf_hub_download

# Suppress warnings
warnings.filterwarnings("ignore")

# Load ensemble model from Hugging Face Hub
def load_model():
    model_path = hf_hub_download(
        repo_id="Zeyadd-Mostaffa/final_ensemble_model",
        filename="final_ensemble_model.pkl"
    )
    model = joblib.load(model_path)
    print("✅ Ensemble model loaded successfully.")
    return model

model = load_model()

# Define prediction function
def predict_employee_status(
    satisfaction_level, last_evaluation, number_project,
    average_monthly_hours, time_spend_company,
    work_accident, promotion_last_5years, salary, department, threshold=0.5
):
    # List of all departments as encoded during training
    departments = [
        'IT', 'RandD', 'accounting', 'hr', 'management',
        'marketing', 'product_mng', 'sales', 'support', 'technical'
    ]

    # One-hot encode department
    department_features = {f"department_{dept}": 0 for dept in departments}
    if department in departments:
        department_features[f"department_{department}"] = 1

    # Interaction features
    satisfaction_evaluation = satisfaction_level * last_evaluation
    work_balance = average_monthly_hours / number_project

    # Construct input DataFrame
    input_data = {
        "satisfaction_level": [satisfaction_level],
        "last_evaluation": [last_evaluation],
        "number_project": [number_project],
        "average_monthly_hours": [average_monthly_hours],
        "time_spend_company": [time_spend_company],
        "Work_accident": [work_accident],
        "promotion_last_5years": [promotion_last_5years],
        "salary": [salary],
        "satisfaction_evaluation": [satisfaction_evaluation],
        "work_balance": [work_balance],
        **department_features
    }

    input_df = pd.DataFrame(input_data)

    try:
        prob = model.predict_proba(input_df)[0][1]
        result = "✅ Employee is likely to quit." if prob >= threshold else "✅ Employee is likely to stay."
        return f"{result} (Probability: {prob:.2%})"
    except Exception as e:
        return f"❌ Prediction error: {str(e)}"

# Gradio Interface
def gradio_interface():
    interface = gr.Interface(
        fn=predict_employee_status,
        inputs=[
            gr.Number(label="Satisfaction Level (0.0 - 1.0)"),
            gr.Number(label="Last Evaluation (0.0 - 1.0)"),
            gr.Number(label="Number of Projects (1 - 10)"),
            gr.Number(label="Average Monthly Hours (80 - 320)"),
            gr.Number(label="Time Spend at Company (Years)"),
            gr.Radio([0, 1], label="Work Accident (0 = No, 1 = Yes)"),
            gr.Radio([0, 1], label="Promotion in Last 5 Years (0 = No, 1 = Yes)"),
            gr.Radio([0, 1, 2], label="Salary (0 = Low, 1 = Medium, 2 = High)"),
            gr.Dropdown(
                ['IT', 'RandD', 'accounting', 'hr', 'management',
                 'marketing', 'product_mng', 'sales', 'support', 'technical'],
                label="Department"
            ),
            gr.Slider(0.1, 0.9, value=0.5, step=0.05, label="Prediction Threshold")
        ],
        outputs="text",
        title="Employee Retention Prediction System (Voting Ensemble)",
        description="Predict whether an employee is likely to stay or quit based on their profile. Supports threshold adjustment.",
        theme="dark"
    )
    interface.launch()

gradio_interface()