app.py

import streamlit as st
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import joblib
from sklearn.model_selection import train_test_split
from matplotlib import rcParams

# Enhanced styling setup
sns.set_style("whitegrid")
plt.style.use('seaborn-v0_8')
rcParams['figure.facecolor'] = '#FAFAFC'
rcParams['axes.facecolor'] = '#FFFFFF'
rcParams['axes.spines.top'] = False
rcParams['axes.spines.right'] = False

# Custom color palette
colors = {
    'primary': '#6366F1',  # Indigo
    'secondary': '#8B5CF6',  # Violet
    'accent': '#06B6D4',  # Cyan
    'success': '#10B981',  # Emerald
    'warning': '#F59E0B',  # Amber
    'bg': '#F8FAFC',  # Slate light
    'card': '#FFFFFF'
}

# ---------- Load data & model ----------
df = pd.read_csv("data/insurance.csv")
@st.cache_resource
def load_model():
    return joblib.load("models/final_insurance_model.pkl")

model = load_model() 

# Hard-code test metrics
MAE = 2538.91
RMSE = 4602.51
R2 = 0.8636

# Pre-split for plot
X_all = df.drop("charges", axis=1)
y_all = df["charges"]
X_train_plot, X_test_plot, y_train_plot, y_test_plot = train_test_split(
    X_all, y_all, test_size=0.2, random_state=42
)

# ---------- Page config ----------
st.set_page_config(
    page_title="Medical Insurance Cost Dashboard",
    page_icon="💊",
    layout="wide"
)

st.title("💊 Medical Insurance Cost Prediction")
st.markdown("---")

# ---------- Tabs ----------
tab_overview, tab_eda, tab_model ,tab_about= st.tabs(["📊 Overview", "🔍 EDA", "🤖 Model & Prediction","ℹ️ About"])

# ===== Overview tab =====
with tab_overview:
    st.subheader("Dataset Overview")

    col1, col2 = st.columns([1, 1])
    with col1:
        st.markdown("**First 5 rows**")
        st.dataframe(df.head(), use_container_width=True)

    with col2:
        st.markdown("**Summary Statistics**")
        st.dataframe(df.describe().round(2), use_container_width=True)

    st.markdown(
        """
        **📋 Feature Descriptions**
        - `age`: Age of insured person (18-64)  
        - `sex`: Gender (male/female)  
        - `bmi`: Body Mass Index (10-60)  
        - `children`: Number of dependents (0-5)  
        - `smoker`: Smoking status (yes/no) ← **Strongest predictor**  
        - `region`: US region (4 categories)  
        - `charges`: Annual medical cost (target)  
        """
    )

# ===== EDA tab =====
with tab_eda:
    # 1. Distribution with gradient fill
    st.subheader("📈 Charges Distribution")
    fig1, ax1 = plt.subplots(figsize=(10, 6))
    sns.histplot(df["charges"], bins=35, kde=True, ax=ax1,
                 color=colors['primary'], alpha=0.7, edgecolor='white')
    ax1.set_xlabel("Annual Medical Charges ($)", fontsize=12, weight='bold')
    ax1.set_ylabel("Frequency", fontsize=12, weight='bold')
    ax1.set_title("Distribution of Medical Insurance Charges",
                  fontsize=16, weight='bold', pad=20)
    plt.tight_layout()
    st.pyplot(fig1)

    st.markdown(
        "🔍 **Key Insight**: Highly **right-skewed distribution** - most pay low-moderate amounts, "
        "few have extremely high costs."
    )

    # 2. Numeric scatter plots
    st.subheader("📊 Numeric Features vs Charges")
    col1, col2 = st.columns(2)

    with col1:
        fig2, ax2 = plt.subplots(figsize=(8, 5))
        scatter = sns.scatterplot(data=df, x="age", y="charges",
                                  alpha=0.7, s=60, color=colors['accent'], ax=ax2)
        ax2.set_title("Age vs Medical Charges", fontsize=14, weight='bold', pad=15)
        ax2.set_xlabel("Age", fontsize=12, weight='bold')
        ax2.set_ylabel("Charges ($)", fontsize=12, weight='bold')
        plt.tight_layout()
        st.pyplot(fig2)
        st.markdown("👴 **Older age → Higher charges** (clear upward trend)")

    with col2:
        fig3, ax3 = plt.subplots(figsize=(8, 5))
        sns.scatterplot(data=df, x="bmi", y="charges",
                        alpha=0.7, s=60, color=colors['warning'], ax=ax3)
        ax3.set_title("BMI vs Medical Charges", fontsize=14, weight='bold', pad=15)
        ax3.set_xlabel("BMI", fontsize=12, weight='bold')
        ax3.set_ylabel("Charges ($)", fontsize=12, weight='bold')
        plt.tight_layout()
        st.pyplot(fig3)
        st.markdown("⚖️ **Higher BMI → Higher charges** (especially BMI > 35)")

    # 3. Categorical boxplots
    st.subheader("🎯 Categorical Impact")
    col3, col4 = st.columns(2)

    with col3:
        fig4, ax4 = plt.subplots(figsize=(8, 5))
        sns.boxplot(data=df, x="smoker", y="charges", ax=ax4,
                    palette=[colors['success'], colors['warning']])
        ax4.set_title("Smoker Status Impact", fontsize=14, weight='bold', pad=15)
        ax4.set_ylabel("Charges ($)", fontsize=12, weight='bold')
        plt.tight_layout()
        st.pyplot(fig4)
        st.markdown("🚬 **Smokers: 4x higher median charges!**")

    with col4:
        fig5, ax5 = plt.subplots(figsize=(8, 5))
        sns.boxplot(data=df, x="region", y="charges", ax=ax5,
                    palette=['#EF4444', '#F59E0B', '#10B981', '#3B82F6'])
        ax5.set_title("Regional Differences", fontsize=14, weight='bold', pad=15)
        ax5.set_ylabel("Charges ($)", fontsize=12, weight='bold')
        plt.tight_layout()
        st.pyplot(fig5)
        st.markdown("🌍 **Regions similar** - smoking dominates geography")

    # 4. Correlation heatmap
    st.subheader("🔗 Feature Correlations")
    fig6, ax6 = plt.subplots(figsize=(7, 5))
    corr = df[["age", "bmi", "children", "charges"]].corr()
    sns.heatmap(corr, annot=True, cmap="RdYlBu_r", fmt=".2f",
                linewidths=1, linecolor='white', cbar_kws={'shrink': 0.8}, ax=ax6)
    ax6.set_title("Correlation Matrix", fontsize=14, weight='bold', pad=20)
    plt.tight_layout()
    st.pyplot(fig6)

    st.markdown("**Top correlations**: Age (0.30) & BMI (0.20) → Charges")

# ===== Model tab =====
with tab_model:
    st.subheader("🎯 Live Prediction")

    # Input form
    col1, col2, col3 = st.columns(3)
    with col1:
        age = st.number_input("👤 Age", min_value=18, max_value=100, value=30)
        children = st.number_input("👨‍👩‍👧‍👦 Children", min_value=0, max_value=5, value=0)
    with col2:
        sex = st.selectbox("⚥️ Sex", ["male", "female"])
        smoker = st.selectbox("🚬 Smoker", ["yes", "no"])
    with col3:
        bmi = st.number_input("⚖️ BMI", min_value=10.0, max_value=60.0, value=25.0, step=0.1)
        region = st.selectbox("📍 Region", ["southeast", "southwest", "northeast", "northwest"])

    if st.button("🔮 Predict Cost", type="primary", use_container_width=True):
        input_df = pd.DataFrame({
            "age": [age], "sex": [sex], "bmi": [bmi],
            "children": [children], "smoker": [smoker], "region": [region]
        })
        user_pred = model.predict(input_df)[0]
        st.success(f"💰 **Estimated Annual Cost: ${user_pred:,.0f}**")
        st.balloons()
    else:
        st.info("📝 Fill inputs above → Click **Predict Cost**")

    st.markdown("---")

    # Performance metrics
    st.subheader("📊 Model Performance")
    colm1, colm2, colm3 = st.columns(3)
    colm1.metric("🎯 MAE", f"${MAE:,.0f}")
    colm2.metric("📈 RMSE", f"${RMSE:,.0f}")
    colm3.metric("⭐ R²", f"{R2:.3f}")

    st.markdown(
        "**✅ Excellent model**: R²=0.86 means 86% of cost variation explained!"
    )

    st.markdown("---")

    # Prediction plot
    st.subheader("🎲 Actual vs Predicted")
    y_test_pred = model.predict(X_test_plot)

    fig7, ax7 = plt.subplots(figsize=(9, 7))
    scatter = ax7.scatter(y_test_plot, y_test_pred, alpha=0.7, s=40,
                          c=y_test_pred - y_test_plot, cmap='RdYlBu_r',
                          edgecolors='white', linewidth=0.5)
    max_val = max(y_test_plot.max(), y_test_pred.max()) * 1.05
    ax7.plot([0, max_val], [0, max_val], "crimson", linewidth=3, alpha=0.9, label="Perfect Prediction")
    ax7.set_xlabel("Actual Charges ($)", fontsize=12, weight='bold')
    ax7.set_ylabel("Predicted Charges ($)", fontsize=12, weight='bold')
    ax7.set_title("Model Accuracy: Actual vs Predicted", fontsize=16, weight='bold', pad=20)
    ax7.legend()
    plt.colorbar(scatter, label="Prediction Error", shrink=0.8)
    plt.tight_layout()
    st.pyplot(fig7)

    st.markdown(
        "🎯 **Points near red line = accurate predictions**. "
        "Color shows error: **blue=underpredict**, **red=overpredict**."
    )
    with tab_about:
        st.subheader("ℹ️ About this app")

        st.markdown(
            """
            This app is a **Medical Insurance Cost Prediction Dashboard** created as a
            personal project to practice data analysis, visualization, and deploying a
            machine learning model to the cloud.

            **Main features**
            - Exploratory Data Analysis of the medical insurance dataset
            - Random Forest model to estimate annual insurance charges
            - Interactive UI where users can input their details and see predictions

            **Author**
            - Name: Pratik Rath
            - Email: pratikrath28@gmail.com
            - LinkedIn: [Pratik Rath](https://www.linkedin.com/in/pratik-rath-02b527394/)
            - GitHub: [pratikrath126](https://github.com/pratikrath126)

            Thank you for checking out this project!
            """,
            unsafe_allow_html=False,
        )