prediction.py

import streamlit as st
import pandas as pd
import pickle
import plotly.express as px


def predict():
    
    
    with open('./model/xgb_tuned.pkl', 'rb') as file_1:
        prediction_model = pickle.load(file_1)

    with open('./model/model_kmeans.pkl', 'rb') as file_2:
        cluster_model = pickle.load(file_2)
    
    #biodata
    st.subheader("Customer Biodata")
    col1, col2, col3 = st.columns(3)
    gender = col1.radio(label="Gender", options=["M", "F"])
    marital = col2.selectbox(label="Marital Status", options=["Single", "Married", "Divorced", "Unkown"])
    relationship = col3.number_input(label="No. of Relationshop", step=1, value=4)
    
    col1, col2, col3 = st.columns(3)
    education = col1.selectbox(label="Education Level", options=["Graduate", "Unedicated", "High School", "College", "Post=Graduate", "Doctorate"])
    income = col2.selectbox(label="Income Bracket", options=["Less than $40K", "$40K - $60K", "$60K - $80K", "$80K - $120K", "$120K+", "Unkown"])
    card = col3.radio(label="Card Type", options=["Blue", "Silver", "Gold", "Platinum"])
    
    #behaviour
    st.subheader("Customer Behaviour")
    
    col1, col2, col3 = st.columns(3)
    active_months = col1.slider(label="Months Inactive", max_value=6, value=2)
    last_contact = col2.slider(label="Last Contact", max_value=6, value=2)
    credit_limit = col3.number_input(label="Credit Limit", step=0.1, value=10834.0)
    
    col1, col2, col3 = st.columns(3)
    revolving_bal = col1.number_input(label="Revolving Balance", step=1, value=0)
    transaction_count = col2.number_input(label="Transaction Count", step=1, value=51)
    transaction_amount = col3.number_input(label="Transaction Amount", step=1, value=2249)
    
    col1, col2, col3 = st.columns(3)
    amount_chng = col1.slider(label="Amount Change",step=0.01,min_value=0.0, max_value=5.0, value=0.522)
    count_change = col2.slider(label="Count Change", step=0.01,min_value=0.0, max_value=5.0, value=0.594)
    util_ratio = col3.slider(label="Utilization Ratio", step=0.01,min_value=0.0, max_value=1.0, value=0.0)
    
    submit = st.button(label="Predict")
    
    if submit:
        data_inf = {
                'Total_Relationship_Count': [relationship],
                'Months_Inactive_12_mon': [active_months],
                'Contacts_Count_12_mon': [last_contact],
                'Credit_Limit': [credit_limit],
                'Total_Revolving_Bal': [revolving_bal],
                'Total_Amt_Chng_Q4_Q1': [amount_chng],
                'Total_Trans_Amt': [transaction_amount],
                'Total_Trans_Ct': [transaction_count],
                'Total_Ct_Chng_Q4_Q1': [count_change],
                'Avg_Utilization_Ratio': [util_ratio],
                'Gender': [gender],
                'Education_Level': [education],
                'Marital_Status': [marital],
                'Income_Category': [income],
                'Card_Category': [card],
        }
        
        data_inf= pd.DataFrame(data_inf)
        
        pred_inf = prediction_model.predict(data_inf)
        if pred_inf == 0:
            pred_inf = "Attrited Customer"
            color = "red"
        else:
            pred_inf = "Existing Customer"
            color = "green"
            
        if pred_inf == "Attrited Customer":
            cluster_inf = cluster_model.predict(data_inf)
        else:
            cluster_inf = None

        if cluster_inf == 0:
            cluster_inf = "Cluster 1 : High Spent Amount, High Usage Frequency"
            recommendation = '<ul><li>1. Rewards and Recognition</li><li>2. Personalized Financial Solution</li><li>3. Financial Education</li></ul>'
        elif cluster_inf == 1:
            cluster_inf = "Cluster 2 : Low Spent Amount, Low Usage Frequency"
            recommendation = '<ul><li>1. Improved Credit Opportunities</li><li>2. Value Propositions</li><li>3. Fee Structure Transparency</li><li>4. Financial Planning Assistance</li></ul>                            '
        
        
        result_html =   """
                        <div style="background-color:#f0f0f0; padding:10px; border-radius:10px">
                            <p style="font-size:16px;"><b>Customer Information:</b></p>
                            <div style="margin-top: 20px;">
                            </div>
                            <p>Customer is predicted to be <span style="color:{color};"><b>{pred_inf}</b></span>, and belongs to <span style="color:blue;"><b>{cluster_inf}</b></span>.</p>
                            <p><b>Here are some recommendations to help reduce churn among customers in corresponding clusters:</b></p>
                            {step}
                        </div>
                        """
        st.markdown(result_html.format(pred_inf=pred_inf, cluster_inf=cluster_inf, color=color, step=recommendation), unsafe_allow_html=True)

def cluster():
    clusters = pd.read_csv('./csv/Cluster.csv')
    bank_df_pca = pd.read_csv('./csv/BankPCA.csv')
    
    colors = {0: 'navy', 1: 'teal'}
    names = {0: 'High Spent Amount (>4K), High Usage Frequency', 
             1: 'Low Spent Amount (<4K), Low Usage Frequency'}
    
    bank_df_pca['color'] = bank_df_pca['label'].map(colors)
    bank_df_pca['name'] = bank_df_pca['label'].map(names)
    
    fig = px.scatter(bank_df_pca, x='x', y='y', color='name', hover_name='name',
                     title='Churn Customer Clustering', width=800, height=400, )
    
    fig.update_traces(marker=dict(size=5))
    fig.update_layout(showlegend=True)
    
    fig.update_layout(height=600)
    st.plotly_chart(fig, use_container_width=True)
    
    
if __name__ == "__main__":
    predict()