streamlit_app.py

import streamlit as st
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
from plotly.subplots import make_subplots
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt

# Set page config
st.set_page_config(
    page_title="Customer Churn Analysis",
    page_icon="📊",
    layout="wide",
    initial_sidebar_state="expanded"
)

@st.cache_data
def load_data():
    """Load and combine the churn datasets"""
    try:
        df1 = pd.read_csv('churn-bigml-20.csv')
        df2 = pd.read_csv('churn-bigml-80.csv')
        

        
        # Add dataset identifier
        df1['Dataset'] = 'Test (20%)'
        df2['Dataset'] = 'Train (80%)'
        
        # Combine datasets
        df_combined = pd.concat([df1, df2], ignore_index=True)
        
        return df1, df2, df_combined
    except Exception as e:
        st.error(f"Error loading data: {str(e)}")
        return None, None, None

def main():
    st.title("📊 Customer Churn Analysis - Exploratory Data Analysis")
    st.markdown("---")
    
    # Load data
    df_test, df_train, df_combined = load_data()
    
    # Check if data loading was successful
    if df_test is None or df_train is None or df_combined is None:
        st.error("Failed to load data. Please check the CSV files exist and are properly formatted.")
        return
    
    # Sidebar
    st.sidebar.title("📋 Analysis Options")
    
    # Dataset selection
    dataset_option = st.sidebar.selectbox(
        "Select Dataset:",
        ["Combined Dataset", "Training Set (80%)", "Test Set (20%)"]
    )
    
    if dataset_option == "Combined Dataset":
        df = df_combined
        st.sidebar.info(f"📈 Total Records: {len(df):,}")
    elif dataset_option == "Training Set (80%)":
        df = df_train
        st.sidebar.info(f"📈 Training Records: {len(df):,}")
    else:
        df = df_test
        st.sidebar.info(f"📈 Test Records: {len(df):,}")
    
    # Analysis sections
    analysis_type = st.sidebar.selectbox(
        "Choose Analysis Type:",
        [
            "📋 Dataset Overview",
            "🎯 Churn Analysis", 
            "📍 Geographic Analysis",
            "📞 Usage Patterns",
            "💰 Revenue Analysis",
            "🔗 Correlation Analysis",
            "📊 Advanced Insights"
        ]
    )
    
    if analysis_type == "📋 Dataset Overview":
        dataset_overview(df)
    elif analysis_type == "🎯 Churn Analysis":
        churn_analysis(df)
    elif analysis_type == "📍 Geographic Analysis":
        geographic_analysis(df)
    elif analysis_type == "📞 Usage Patterns":
        usage_patterns(df)
    elif analysis_type == "💰 Revenue Analysis":
        revenue_analysis(df)
    elif analysis_type == "🔗 Correlation Analysis":
        correlation_analysis(df)
    elif analysis_type == "📊 Advanced Insights":
        advanced_insights(df)

def dataset_overview(df):
    st.header("📋 Dataset Overview")
    
    # Dataset description
    st.markdown("""
    ### 📞 About This Dataset
    
    This is a **telecommunications customer churn dataset** that contains information about customers of a telecom company and whether they churned (cancelled their service) or were retained. The dataset is commonly used for predictive modeling to identify customers at risk of churning.
    
    **Key Characteristics:**
    - **Domain**: Telecommunications industry
    - **Target Variable**: `Churn` (True/False) - indicates if customer cancelled service
    - **Time Period**: Historical customer data with usage patterns and service details
    - **Geographic Coverage**: Multiple US states (51 unique states)
    - **Use Case**: Customer retention analysis, churn prediction modeling, and business intelligence
    
    **Feature Categories:**
    - 🏠 **Demographics**: State, account length, area code
    - 📋 **Service Plans**: International calling plan, voice mail plan
    - 📞 **Usage Patterns**: Day/evening/night/international minutes, calls, and charges
    - 🎧 **Service Interactions**: Customer service call frequency
    - 💰 **Billing**: Detailed breakdown of charges by time period and service type
    
    This dataset enables analysis of customer behavior patterns, identification of churn risk factors, and development of retention strategies.
    """)
    
    st.markdown("---")
    
    # Basic info
    col1, col2, col3, col4 = st.columns(4)
    
    with col1:
        st.metric("Total Records", f"{len(df):,}")
    with col2:
        st.metric("Total Features", len(df.columns))
    with col3:
        if 'Churn' in df.columns:
            churn_rate = (df['Churn'] == True).mean() * 100
            st.metric("Churn Rate", f"{churn_rate:.1f}%")
        else:
            st.metric("Churn Rate", "N/A")
    with col4:
        if 'State' in df.columns:
            st.metric("Unique States", df['State'].nunique())
        else:
            st.metric("Unique States", "N/A")
    
    # Dataset structure
    st.subheader("📊 Dataset Structure")
    
    col1, col2 = st.columns(2)
    
    with col1:
        st.write("**Data Types:**")
        data_types = df.dtypes.value_counts()
        # Convert index to string to avoid JSON serialization issues
        fig = px.pie(
            values=data_types.values,
            names=[str(dtype) for dtype in data_types.index],
            title="Distribution of Data Types"
        )
        st.plotly_chart(fig, use_container_width=True)
    
    with col2:
        st.write("**Missing Values:**")
        missing_data = df.isnull().sum()
        if missing_data.sum() == 0:
            st.success("✅ No missing values found!")
        else:
            st.write(missing_data[missing_data > 0])
    
    # Sample data
    st.subheader("🔍 Sample Data")
    st.dataframe(df.head(10), use_container_width=True)
    

    
    # Statistical summary
    st.subheader("📈 Statistical Summary")
    st.dataframe(df.describe(), use_container_width=True)

def churn_analysis(df):
    st.header("🎯 Churn Analysis")
    
    # Check if Churn column exists
    if 'Churn' not in df.columns:
        st.error("'Churn' column not found in the dataset. Please check your data.")
        return
    
    # Churn distribution
    col1, col2 = st.columns(2)
    
    with col1:
        churn_counts = df['Churn'].value_counts()
        fig = px.pie(
            values=churn_counts.values,
            names=['Retained', 'Churned'],
            title="Overall Churn Distribution",
            color_discrete_sequence=['lightgreen', 'lightcoral']
        )
        st.plotly_chart(fig, use_container_width=True)
    
    with col2:
        # Churn by categorical variables
        categorical_vars = ['International plan', 'Voice mail plan']
        selected_var = st.selectbox("Select categorical variable:", categorical_vars)
        
        churn_by_cat = df.groupby([selected_var, 'Churn']).size().unstack()
        churn_rate_cat = df.groupby(selected_var)['Churn'].mean() * 100
        
        fig = px.bar(
            x=churn_rate_cat.index,
            y=churn_rate_cat.values,
            title=f"Churn Rate by {selected_var}",
            labels={'x': selected_var, 'y': 'Churn Rate (%)'},
            color=churn_rate_cat.values,
            color_continuous_scale='Reds'
        )
        st.plotly_chart(fig, use_container_width=True)
    
    # Churn by numerical variables
    st.subheader("📊 Churn Analysis by Numerical Features")
    
    numerical_vars = [
        'Account length', 'Total day minutes', 'Total day calls', 'Total day charge',
        'Total eve minutes', 'Total eve calls', 'Total eve charge',
        'Total night minutes', 'Total night calls', 'Total night charge',
        'Total intl minutes', 'Total intl calls', 'Total intl charge',
        'Customer service calls'
    ]
    
    selected_num_var = st.selectbox("Select numerical variable:", numerical_vars)
    
    col1, col2 = st.columns(2)
    
    with col1:
        # Box plot
        fig = px.box(
            df, x='Churn', y=selected_num_var,
            title=f"{selected_num_var} Distribution by Churn Status",
            color='Churn',
            color_discrete_sequence=['lightgreen', 'lightcoral']
        )
        st.plotly_chart(fig, use_container_width=True)
    
    with col2:
        # Histogram
        fig = px.histogram(
            df, x=selected_num_var, color='Churn',
            title=f"{selected_num_var} Distribution",
            marginal="box",
            color_discrete_sequence=['lightgreen', 'lightcoral']
        )
        st.plotly_chart(fig, use_container_width=True)

def geographic_analysis(df):
    st.header("📍 Geographic Analysis")
    
    # Check required columns
    required_cols = ['State', 'Churn']
    missing_cols = [col for col in required_cols if col not in df.columns]
    if missing_cols:
        st.error(f"Required columns missing: {missing_cols}")
        return
    
    # State-wise analysis
    state_analysis = df.groupby('State').agg({
        'Churn': ['count', 'sum', 'mean'],
        'Total day charge': 'mean',
        'Total eve charge': 'mean',
        'Total night charge': 'mean',
        'Total intl charge': 'mean'
    }).round(2)
    
    state_analysis.columns = ['Total_Customers', 'Churned_Customers', 'Churn_Rate', 
                             'Avg_Day_Charge', 'Avg_Eve_Charge', 'Avg_Night_Charge', 'Avg_Intl_Charge']
    state_analysis = state_analysis.reset_index()
    
    col1, col2 = st.columns(2)
    
    with col1:
        # Churn rate by state
        fig = px.choropleth(
            state_analysis,
            locations='State',
            color='Churn_Rate',
            hover_name='State',
            hover_data=['Total_Customers', 'Churned_Customers'],
            locationmode='USA-states',
            title="Churn Rate by State",
            color_continuous_scale='Reds'
        )
        fig.update_layout(geo_scope="usa")
        st.plotly_chart(fig, use_container_width=True)
    
    with col2:
        # Top 10 states by churn rate
        top_churn_states = state_analysis.nlargest(10, 'Churn_Rate')
        fig = px.bar(
            top_churn_states,
            x='Churn_Rate',
            y='State',
            orientation='h',
            title="Top 10 States by Churn Rate",
            color='Churn_Rate',
            color_continuous_scale='Reds'
        )
        st.plotly_chart(fig, use_container_width=True)
    
    # Area code analysis
    st.subheader("📞 Area Code Analysis")
    
    area_code_analysis = df.groupby('Area code').agg({
        'Churn': ['count', 'mean'],
        'Total day charge': 'mean'
    }).round(2)
    
    area_code_analysis.columns = ['Total_Customers', 'Churn_Rate', 'Avg_Day_Charge']
    area_code_analysis = area_code_analysis.reset_index()
    
    fig = px.bar(
        area_code_analysis,
        x='Area code',
        y='Churn_Rate',
        title="Churn Rate by Area Code",
        color='Churn_Rate',
        color_continuous_scale='Reds',
        hover_data=['Total_Customers']
    )
    st.plotly_chart(fig, use_container_width=True)

def usage_patterns(df):
    st.header("📞 Usage Patterns Analysis")
    
    # Check if Churn column exists
    if 'Churn' not in df.columns:
        st.error("'Churn' column not found in the dataset.")
        return
    
    # Time-based usage analysis
    usage_metrics = ['Total day minutes', 'Total eve minutes', 'Total night minutes', 'Total intl minutes']
    
    col1, col2 = st.columns(2)
    
    with col1:
        # Usage patterns by churn
        usage_by_churn = df.groupby('Churn')[usage_metrics].mean()
        
        fig = go.Figure()
        for metric in usage_metrics:
            fig.add_trace(go.Bar(
                name=metric.replace('Total ', '').replace(' minutes', ''),
                x=['Retained', 'Churned'],
                y=[usage_by_churn.loc[False, metric], usage_by_churn.loc[True, metric]]
            ))
        
        fig.update_layout(
            title="Average Usage Patterns by Churn Status",
            xaxis_title="Customer Status",
            yaxis_title="Average Minutes",
            barmode='group'
        )
        st.plotly_chart(fig, use_container_width=True)
    
    with col2:
        # Call frequency analysis
        call_metrics = ['Total day calls', 'Total eve calls', 'Total night calls', 'Total intl calls']
        call_by_churn = df.groupby('Churn')[call_metrics].mean()
        
        fig = go.Figure()
        for metric in call_metrics:
            fig.add_trace(go.Bar(
                name=metric.replace('Total ', '').replace(' calls', ''),
                x=['Retained', 'Churned'],
                y=[call_by_churn.loc[False, metric], call_by_churn.loc[True, metric]]
            ))
        
        fig.update_layout(
            title="Average Call Frequency by Churn Status",
            xaxis_title="Customer Status",
            yaxis_title="Average Number of Calls",
            barmode='group'
        )
        st.plotly_chart(fig, use_container_width=True)
    
    # Customer service calls analysis
    st.subheader("🎧 Customer Service Analysis")
    
    col1, col2 = st.columns(2)
    
    with col1:
        cs_calls_churn = df.groupby('Customer service calls')['Churn'].agg(['count', 'sum', 'mean']).reset_index()
        cs_calls_churn['churn_rate'] = cs_calls_churn['mean'] * 100
        
        fig = px.bar(
            cs_calls_churn,
            x='Customer service calls',
            y='churn_rate',
            title="Churn Rate by Number of Customer Service Calls",
            color='churn_rate',
            color_continuous_scale='Reds',
            hover_data=['count']
        )
        st.plotly_chart(fig, use_container_width=True)
    
    with col2:
        # Distribution of customer service calls
        fig = px.histogram(
            df,
            x='Customer service calls',
            color='Churn',
            title="Distribution of Customer Service Calls",
            color_discrete_sequence=['lightgreen', 'lightcoral']
        )
        st.plotly_chart(fig, use_container_width=True)

def revenue_analysis(df):
    st.header("💰 Revenue Analysis")
    
    # Check if Churn column exists
    if 'Churn' not in df.columns:
        st.error("'Churn' column not found in the dataset.")
        return
    
    # Calculate total revenue per customer
    df['Total_Revenue'] = (df['Total day charge'] + df['Total eve charge'] + 
                          df['Total night charge'] + df['Total intl charge'])
    
    col1, col2 = st.columns(2)
    
    with col1:
        # Revenue by churn status
        revenue_by_churn = df.groupby('Churn')['Total_Revenue'].agg(['mean', 'median', 'std']).round(2)
        
        fig = px.bar(
            x=['Retained', 'Churned'],
            y=[revenue_by_churn.loc[False, 'mean'], revenue_by_churn.loc[True, 'mean']],
            title="Average Revenue by Churn Status",
            color=['Retained', 'Churned'],
            color_discrete_sequence=['lightgreen', 'lightcoral']
        )
        fig.update_layout(yaxis_title="Average Revenue ($)")
        st.plotly_chart(fig, use_container_width=True)
    
    with col2:
        # Revenue distribution
        fig = px.box(
            df,
            x='Churn',
            y='Total_Revenue',
            title="Revenue Distribution by Churn Status",
            color='Churn',
            color_discrete_sequence=['lightgreen', 'lightcoral']
        )
        st.plotly_chart(fig, use_container_width=True)
    
    # Revenue breakdown analysis
    st.subheader("📊 Revenue Breakdown")
    
    revenue_components = ['Total day charge', 'Total eve charge', 'Total night charge', 'Total intl charge']
    
    col1, col2 = st.columns(2)
    
    with col1:
        # Average revenue components
        avg_components = df[revenue_components].mean()
        
        fig = px.pie(
            values=avg_components.values,
            names=[comp.replace('Total ', '').replace(' charge', '') for comp in revenue_components],
            title="Average Revenue Composition"
        )
        st.plotly_chart(fig, use_container_width=True)
    
    with col2:
        # Revenue components by churn
        components_by_churn = df.groupby('Churn')[revenue_components].mean()
        
        fig = go.Figure()
        for component in revenue_components:
            fig.add_trace(go.Bar(
                name=component.replace('Total ', '').replace(' charge', ''),
                x=['Retained', 'Churned'],
                y=[components_by_churn.loc[False, component], components_by_churn.loc[True, component]]
            ))
        
        fig.update_layout(
            title="Revenue Components by Churn Status",
            xaxis_title="Customer Status",
            yaxis_title="Average Charge ($)",
            barmode='group'
        )
        st.plotly_chart(fig, use_container_width=True)

def correlation_analysis(df):
    st.header("🔗 Correlation Analysis")
    
    # Check if Churn column exists
    if 'Churn' not in df.columns:
        st.error("'Churn' column not found in the dataset.")
        return
    
    # Select numerical columns for correlation
    numerical_cols = df.select_dtypes(include=[np.number]).columns.tolist()
    if 'Dataset' in numerical_cols:
        numerical_cols.remove('Dataset')
    
    # Convert boolean to numerical for correlation
    df_corr = df.copy()
    df_corr['Churn'] = df_corr['Churn'].astype(int)
    df_corr['International plan'] = (df_corr['International plan'] == 'Yes').astype(int)
    df_corr['Voice mail plan'] = (df_corr['Voice mail plan'] == 'Yes').astype(int)
    
    # Add converted columns to numerical_cols for correlation
    correlation_cols = numerical_cols.copy()
    if 'Churn' not in correlation_cols:
        correlation_cols.append('Churn')
    if 'International plan' in df_corr.columns and 'International plan' not in correlation_cols:
        correlation_cols.append('International plan')
    if 'Voice mail plan' in df_corr.columns and 'Voice mail plan' not in correlation_cols:
        correlation_cols.append('Voice mail plan')
    
    # Calculate correlation matrix
    corr_matrix = df_corr[correlation_cols].corr()
    
    col1, col2 = st.columns(2)
    
    with col1:
        # Correlation heatmap
        fig = px.imshow(
            corr_matrix,
            title="Feature Correlation Heatmap",
            color_continuous_scale='RdBu_r',
            aspect="auto"
        )
        fig.update_layout(width=600, height=600)
        st.plotly_chart(fig, use_container_width=True)
    
    with col2:
        # Correlation with churn
        if 'Churn' in corr_matrix.columns:
            churn_corr = corr_matrix['Churn'].abs().sort_values(ascending=False)[1:]  # Exclude self-correlation
            
            fig = px.bar(
                x=churn_corr.values,
                y=churn_corr.index,
                orientation='h',
                title="Features Most Correlated with Churn",
                color=churn_corr.values,
                color_continuous_scale='Reds'
            )
            st.plotly_chart(fig, use_container_width=True)
        else:
            st.error("Churn column not found in correlation matrix.")
    
    # Top correlations
    st.subheader("🔍 Key Correlations")
    
    if 'Churn' in corr_matrix.columns:
        # Find top positive and negative correlations with churn
        churn_corr_full = corr_matrix['Churn'].sort_values(ascending=False)
        
        col1, col2 = st.columns(2)
        
        with col1:
            st.write("**Top Positive Correlations with Churn:**")
            top_positive = churn_corr_full[churn_corr_full > 0][1:6]  # Top 5, excluding self
            if len(top_positive) > 0:
                for feature, corr in top_positive.items():
                    st.write(f"• {feature}: {corr:.3f}")
            else:
                st.write("No positive correlations found.")
        
        with col2:
            st.write("**Top Negative Correlations with Churn:**")
            top_negative = churn_corr_full[churn_corr_full < 0][-5:]  # Bottom 5
            if len(top_negative) > 0:
                for feature, corr in top_negative.items():
                    st.write(f"• {feature}: {corr:.3f}")
            else:
                st.write("No negative correlations found.")
    else:
        st.warning("Cannot display correlation insights without Churn column in correlation matrix.")

def advanced_insights(df):
    st.header("📊 Advanced Insights")
    
    # Check if Churn column exists
    if 'Churn' not in df.columns:
        st.error("'Churn' column not found in the dataset.")
        return
    
    # Customer segments analysis
    st.subheader("👥 Customer Segmentation")
    
    # Calculate total usage and revenue
    df['Total_Usage'] = (df['Total day minutes'] + df['Total eve minutes'] + 
                        df['Total night minutes'] + df['Total intl minutes'])
    df['Total_Revenue'] = (df['Total day charge'] + df['Total eve charge'] + 
                          df['Total night charge'] + df['Total intl charge'])
    
    # Create usage vs revenue scatter plot
    fig = px.scatter(
        df,
        x='Total_Usage',
        y='Total_Revenue',
        color='Churn',
        title="Customer Segmentation: Usage vs Revenue",
        labels={'Total_Usage': 'Total Usage (minutes)', 'Total_Revenue': 'Total Revenue ($)'},
        color_discrete_sequence=['lightgreen', 'lightcoral'],
        opacity=0.6
    )
    st.plotly_chart(fig, use_container_width=True)
    
    # High-risk customer identification
    st.subheader("⚠️ High-Risk Customer Analysis")
    
    # Identify patterns in churned customers
    churned_customers = df[df['Churn'] == True]
    
    col1, col2 = st.columns(2)
    
    with col1:
        # Customer service calls for churned customers
        cs_calls_dist = churned_customers['Customer service calls'].value_counts().sort_index()
        
        fig = px.bar(
            x=cs_calls_dist.index,
            y=cs_calls_dist.values,
            title="Customer Service Calls Distribution (Churned Customers)",
            labels={'x': 'Number of CS Calls', 'y': 'Number of Customers'},
            color=cs_calls_dist.values,
            color_continuous_scale='Reds'
        )
        st.plotly_chart(fig, use_container_width=True)
    
    with col2:
        # Account length distribution for churned customers
        fig = px.histogram(
            churned_customers,
            x='Account length',
            title="Account Length Distribution (Churned Customers)",
            color_discrete_sequence=['lightcoral']
        )
        st.plotly_chart(fig, use_container_width=True)
    
    # Plan analysis
    st.subheader("📋 Plan Analysis")
    
    plan_analysis = df.groupby(['International plan', 'Voice mail plan']).agg({
        'Churn': ['count', 'sum', 'mean']
    }).round(3)
    
    plan_analysis.columns = ['Total_Customers', 'Churned_Customers', 'Churn_Rate']
    plan_analysis = plan_analysis.reset_index()
    plan_analysis['Plan_Combination'] = (plan_analysis['International plan'] + ' Intl, ' + 
                                        plan_analysis['Voice mail plan'] + ' VM')
    
    fig = px.bar(
        plan_analysis,
        x='Plan_Combination',
        y='Churn_Rate',
        title="Churn Rate by Plan Combinations",
        color='Churn_Rate',
        color_continuous_scale='Reds',
        hover_data=['Total_Customers']
    )
    fig.update_xaxes(tickangle=45)
    st.plotly_chart(fig, use_container_width=True)
    
    # Key insights summary
    st.subheader("💡 Key Insights Summary")
    
    try:
        insights = []
        
        # Overall churn rate
        if 'Churn' in df.columns:
            churn_rate = (df['Churn'] == True).mean() * 100
            insights.append(f"📈 Overall churn rate: {churn_rate:.1f}%")
        
        # Customer service calls analysis
        if 'Customer service calls' in df.columns and 'Churn' in df.columns:
            high_service_calls = df[df['Customer service calls'] >= 4]
            if len(high_service_calls) > 0:
                high_cs_churn_rate = high_service_calls['Churn'].mean() * 100
                insights.append(f"📞 Customers with 4+ service calls have {high_cs_churn_rate:.1f}% churn rate")
        
        # International plan analysis
        if 'International plan' in df.columns and 'Churn' in df.columns:
            intl_customers = df[df['International plan'] == 'Yes']
            if len(intl_customers) > 0:
                intl_churn_rate = intl_customers['Churn'].mean() * 100
                insights.append(f"🌍 International plan customers have {intl_churn_rate:.1f}% churn rate")
        
        # Voice mail plan analysis
        if 'Voice mail plan' in df.columns and 'Churn' in df.columns:
            vm_customers = df[df['Voice mail plan'] == 'Yes']
            if len(vm_customers) > 0:
                vm_churn_rate = vm_customers['Churn'].mean() * 100
                insights.append(f"📧 Voice mail plan customers have {vm_churn_rate:.1f}% churn rate")
        
        # Revenue analysis
        if 'Total_Revenue' in df.columns and 'Churn' in df.columns:
            churned_customers = df[df['Churn'] == True]
            if len(churned_customers) > 0:
                avg_revenue_churned = churned_customers['Total_Revenue'].mean()
                insights.append(f"💰 Average revenue per churned customer: ${avg_revenue_churned:.2f}")
        
        # Account length analysis
        if 'Account length' in df.columns and 'Churn' in df.columns:
            churned_customers = df[df['Churn'] == True]
            if len(churned_customers) > 0:
                avg_account_length = churned_customers['Account length'].mean()
                insights.append(f"⏰ Average account length of churned customers: {avg_account_length:.0f} days")
        
        # Display insights
        for insight in insights:
            st.info(insight)
            
        if not insights:
            st.warning("No insights could be generated due to missing required columns.")
            
    except Exception as e:
        st.error(f"Error generating insights: {str(e)}")
        st.warning("Please check that all required columns are present in your dataset.")

if __name__ == "__main__":
    main()