app.py

import io
import pickle
import streamlit as st
import joblib
import shap
import pandas as pd
import matplotlib.pyplot as plt



with open('lgb1_model.pkl', 'rb') as f:
    lgb1 = pickle.load(f)

categorical_features = joblib.load("categorical_features.joblib")
encoder = joblib.load("encoder.joblib")


option = st.sidebar.selectbox("Which dashboard?", ( "Model information", "Stroke prediction","Information about training data"))
st.title(option)

def get_pred():
    """
    Function to display the stroke probability calculator and Shap force plot.
    """
    st.header("Stroke probability calculator ")

    
    gender = st.selectbox("Select gender: ", ["Male", "Female", 'Other'])
    work_type = st.selectbox("Work type: ", ["Private", "Self_employed", 'children', 'Govt_job', 'Never_worked'])
    residence_status = st.selectbox("Residence status: ", ["Urban", "Rural"])
    smoking_status = st.selectbox("Smoking status: ", ["Unknown", "formerly smoked", 'never smoked', 'smokes'])
    age = st.slider("Input age: ", 0, 120)
    hypertension = st.select_slider("Do you have hypertension: ", [0, 1])
    heart_disease = st.select_slider("Do you have heart disease: ", [0, 1])
    ever_married = st.select_slider("Have you ever married? ", [0, 1])
    avg_glucosis_lvl = st.slider("Average glucosis level: ", 50, 280)
    bmi = st.slider("Input Bmi: ", 10, 100)

    
    data = {
        "gender": gender,
        "work_type": work_type,
        "Residence_type": residence_status,
        "smoking_status": smoking_status,
        "age": age,
        "hypertension": hypertension,
        "heart_disease": heart_disease,
        "ever_married": ever_married,
        "avg_glucose_level": avg_glucosis_lvl,
        "bmi": bmi
    }


    if st.button("Predict"):
        X = pd.DataFrame([data])
        
        encoded_features = encoder.transform(X[categorical_features])

        feature_names = encoder.get_feature_names_out(input_features=categorical_features)

        encoded_df = pd.DataFrame(encoded_features, columns=feature_names)
        
        X_encoded = pd.concat([X.drop(columns=categorical_features), encoded_df], axis=1)

        prediction_proba = lgb1.predict_proba(X_encoded)

        explainer = shap.TreeExplainer(lgb1)
        
        shap_values = explainer.shap_values(X_encoded)

        probability = prediction_proba[0, 1]  # Assuming binary classification
        st.subheader(f"The predicted probability of stroke is {probability}.")
        st.subheader("IF you see result , higher than 0.3, we advice you to see a doctor")
        st.header("Shap forceplot")
        st.subheader("Features values impact on model made prediction")

        shap.force_plot(explainer.expected_value[1], shap_values[1], features=X_encoded.iloc[0, :], matplotlib=True)


        buf = io.BytesIO()
        plt.savefig(buf, format="png", dpi=800)
        buf.seek(0)


        st.image(buf, width=1100)


        shap.summary_plot(shap_values[1], X_encoded)


        shap_interaction_values = explainer.shap_interaction_values(X_encoded)
        shap.summary_plot(shap_interaction_values, X_encoded)


if option == "Information about training data":
    st.header("Stroke Prediction Dataset")
    st.subheader("According to the World Health Organization (WHO), stroke is the 2nd leading cause of death globally, responsible for approximately 11% of total deaths. This dataset is used to predict whether a patient is likely to get a stroke based on input parameters like gender, age, various diseases, and smoking status. Each row in the data provides relevant information about the patient.")
    st.markdown("### Disclaimer: This project is made from data collected from one American hospital. For this model to be more relevant in predicting health risks, it should ideally be trained on data specific to your population.")
    
    st.subheader("Stroke dataset has 5110 records and 12 features.")
    st.subheader("Ratio of stroke between respondents.")
    st.image(r'Stroke_between_respondents.png')
    
    st.markdown("### 4.9% of all respondents had a stroke.")
    
    st.subheader("Fun fact:")
    st.markdown("People who have formerly smoked have the highest stroke risk.")
    st.image(r'smoking_status.png')
    
    st.subheader("Age importance in risk of Stroke:")
    st.image(r'Stroke_age.png')
    
    st.markdown("Stroke risk increases significantly after 40 years of age.")
    
    st.subheader("Correlation between features:")
    st.image(r'Correlation.png')
    
    st.subheader("Feature Importance (SHAP values) affecting Stroke prediction:")
    st.image(r'Shap_Values.png')
    

    
if option == "Stroke prediction":
    get_pred()
    st.subheader("Disclaimer: This project is made out of one American hospital data. For this model to be more relevant to predict your health, it has to bee trained on your population data")

if option == "Model information":
    st.header("Light gradient boosting model")
    st.subheader("First tree of light gradient boosting model and how it makes decisions")
    st.image(r'lgbm_tree.png')
    



    st.subheader("Shap values visualization of how features contribute to model prediction")
    st.image(r'lgbm_model_shap_evaluation.png')
    st.subheader("Confusion matrix of predictions made by a Light GradientBoost Classifier and it's ROC_Curve")
    
    col1, col2 = st.columns(2)  

    with col1:
        st.image(r'Confusion_matrix.png')

    with col2:
        st.image(r'Roc_Curve.png')

    
    st.markdown("### Model from 28 stroke cases in a test set, identified bad 1 case")
    st.markdown("### Model is trained for a recall, because it's better to send a person to visit other doctor, that's why model identified from 182 pacients with stroke , 154 bad, they didin't have a stroke")