import streamlit as st import pandas as pd import seaborn as sns import matplotlib.pyplot as plt import plotly.express as px import numpy as np import xgboost as xgb import os from sklearn.model_selection import train_test_split from sklearn.metrics import accuracy_score, classification_report, confusion_matrix # Set page configuration at the very top st.set_page_config(page_title="Healthcare Dashboard", layout="wide", page_icon="💡") # Define human-readable labels for prediction outcomes OUTCOME_MAP = { 0: "Patient recovered and went home", 1: "Transferred to another hospital or care center", 2: "Transferred to a rehab center", 3: "Left the hospital early without approval", 4: "Passed away or had a very serious event", } # Function to load the model def load_model(): try: model = xgb.XGBClassifier() model.load_model("xgboost_patient_model.json") return model except Exception as e: st.error(f"Error loading model: {e}") return None # Ensure data matches model's feature requirements def preprocess_data(data, model_features): try: data = data.apply(pd.to_numeric, errors='coerce') missing_features = [f for f in model_features if f not in data.columns] extra_features = [f for f in data.columns if f not in model_features] if missing_features: st.error(f"❌ Missing required features: {missing_features}") return None if extra_features: st.warning(f"⚠️ Extra features in uploaded data: {extra_features}") return data[model_features] except Exception as e: st.error(f"Data preprocessing error: {e}") return None # Predict patient outcomes def predict_outcome(model, data): if model is None: return None, None, None, None actual_target = data.pop("target") if "target" in data.columns else None try: model_features = model.get_booster().feature_names data = preprocess_data(data, model_features) if data is None: return None, None, None, None predictions = model.predict(data) # Convert numerical predictions to human-readable labels mapped_predictions = [OUTCOME_MAP[pred] for pred in predictions] actual_labels = [OUTCOME_MAP[actual] for actual in actual_target] if actual_target is not None else ["N/A"] * len(predictions) # Debugging information if actual_target is not None: correct_predictions = (predictions == actual_target).sum() total_predictions = len(actual_target) accuracy = (correct_predictions / total_predictions) * 100 st.write(f"✅ Correct Predictions: {correct_predictions}/{total_predictions}") st.write(f"📊 Model Accuracy: **{accuracy:.2f}%**") return actual_target, predictions, mapped_predictions, actual_labels except Exception as e: st.error(f"Prediction error: {e}") return None, None, None, None # Load the data file_path = 'final_cleaned_patient_data.csv' try: df = pd.read_csv(file_path) except Exception as e: st.error(f"Error loading data: {e}") df = pd.DataFrame() # Create empty DataFrame if file doesn't exist # Sidebar navigation st.sidebar.title('Healthcare Data Dashboard') # Team Members Section st.sidebar.markdown("### 🏆 Team Members:") team_members = [ "1. R. Sai Somnath", "2. S. Sreevardhan", "3. S. Mohammad Basha", "4. V. Hussain Basha", "5. P. Charles" ] for member in team_members: st.sidebar.text(member) # Add a divider st.sidebar.markdown("---") # Section navigation - Added two new sections option = st.sidebar.selectbox('Choose a section', [ 'Data Overview', 'Data Visualization', 'Interactive Reports', 'Correlation Analysis', 'Data Insights', 'Patient Outcome Prediction', 'Batch Prediction', 'Model Performance' ]) # Apply a Streamlit theme with a dark background for a modern look st.markdown(""" """, unsafe_allow_html=True) # Data Overview Section if option == 'Data Overview': st.title('📊 Data Overview') st.write(df.head()) st.write(f"Dataset Shape: {df.shape}") st.write(f"Column Names: {df.columns.tolist()}") st.write("Basic Statistical Overview:") st.write(df.describe()) if st.checkbox('Show Missing Values'): st.write(df.isnull().sum()) # Data Visualization Section elif option == 'Data Visualization': st.title('📈 Data Visualization') column = st.selectbox('Select Column for Visualization', df.columns) plot_type = st.radio('Choose plot type', ['Histogram', 'Boxplot', 'Violin Plot', 'Scatter Plot', 'Line Plot', 'Animated Plot']) if plot_type == 'Animated Plot': time_col = st.selectbox('Select Time Column (if applicable)', df.columns) fig = px.scatter(df, x=column, y=column, animation_frame=time_col, size_max=60) elif plot_type == 'Histogram': fig = px.histogram(df, x=column, marginal='box', nbins=30) elif plot_type == 'Boxplot': fig = px.box(df, y=column) elif plot_type == 'Violin Plot': fig = px.violin(df, y=column, box=True, points='all') elif plot_type == 'Scatter Plot': x_col = st.selectbox('Select X axis', df.columns) fig = px.scatter(df, x=x_col, y=column, color=column) elif plot_type == 'Line Plot': x_col = st.selectbox('Select X axis for Line Plot', df.columns) fig = px.line(df, x=x_col, y=column) st.plotly_chart(fig) # Correlation Analysis Section elif option == 'Correlation Analysis': st.title('🔎 Correlation Analysis') corr_matrix = df.corr() fig, ax = plt.subplots(figsize=(12, 8)) sns.heatmap(corr_matrix, annot=True, cmap='coolwarm', fmt='.2f', ax=ax) st.pyplot(fig) # Interactive Reports Section elif option == 'Interactive Reports': st.title('📂 Interactive Reports') st.write("Filter and explore the data.") selected_columns = st.multiselect('Select columns to display', df.columns) st.dataframe(df[selected_columns] if selected_columns else df) st.write("Filter the Data:") filter_column = st.selectbox('Select column to filter by', df.columns) filter_value = st.text_input('Enter filter value') if filter_value: filtered_data = df[df[filter_column].astype(str).str.contains(filter_value, case=False)] st.write(filtered_data) # Download option csv_data = filtered_data.to_csv(index=False).encode('utf-8') st.download_button(label='Download Filtered Data as CSV', data=csv_data, file_name='filtered_data.csv', mime='text/csv') # Data Insights Section elif option == 'Data Insights': st.title('🧠 Data Insights') st.write("Gain insights into the data using various metrics.") st.write("Total Unique Values per Column:") st.write(df.nunique()) st.write("Top 5 Frequent Values for Each Column:") for col in df.columns: st.write(f"{col}: {df[col].value_counts().head(5)}") # Patient Outcome Prediction Section elif option == 'Patient Outcome Prediction': st.title('🤖 Patient Outcome Prediction') # Load the pre-trained model model = load_model() if model is not None: st.success("✅ Pre-trained model loaded successfully!") # Define class descriptions class_descriptions = { 0: "Patient recovered and went home", 1: "Patient transferred to another hospital", 2: "Patient moved to rehab facility", 3: "Patient left against medical advice", 4: "Patient deceased or serious outcome" } # Display target class distribution if target column exists target_column = 'target' if target_column in df.columns: st.subheader("Target Class Distribution") target_counts = df[target_column].value_counts().reset_index() target_counts.columns = ['Class', 'Count'] target_counts['Description'] = target_counts['Class'].map(class_descriptions) st.write(target_counts) fig = px.pie(target_counts, values='Count', names='Description', title='Target Class Distribution') st.plotly_chart(fig) # Prediction interface st.subheader("Make Predictions") st.write("Enter values for the features to predict the patient outcome:") # Create a more interactive UI for prediction with all input values col1, col2, col3 = st.columns(3) # Create input fields for all required features input_values = {} with col1: input_values['age'] = st.number_input("Age", min_value=0, max_value=120, value=51) input_values['gender'] = st.selectbox("Gender", [0, 1], index=1, format_func=lambda x: "Male" if x == 0 else "Female") input_values['previous_hospitalizations'] = st.number_input("Previous Hospitalizations", min_value=0, value=4) input_values['heart_rate'] = st.number_input("Heart Rate", min_value=30, max_value=200, value=63) input_values['respiratory_rate'] = st.number_input("Respiratory Rate", min_value=5, max_value=60, value=16) input_values['blood_pressure_sys'] = st.number_input("Blood Pressure (Systolic)", min_value=50, max_value=250, value=86) input_values['blood_pressure_dia'] = st.number_input("Blood Pressure (Diastolic)", min_value=30, max_value=150, value=58) input_values['temperature'] = st.number_input("Temperature (°C)", min_value=35.0, max_value=42.0, value=35.86, step=0.1) input_values['wbc_count'] = st.number_input("WBC Count", min_value=0.0, max_value=50.0, value=7.15, step=0.1) input_values['creatinine'] = st.number_input("Creatinine", min_value=0.1, max_value=10.0, value=2.93, step=0.1) with col2: input_values['bilirubin'] = st.number_input("Bilirubin", min_value=0.1, max_value=30.0, value=1.72, step=0.1) input_values['glucose'] = st.number_input("Glucose", min_value=40, max_value=500, value=137) input_values['bun'] = st.number_input("BUN", min_value=5, max_value=150, value=36) input_values['pH'] = st.number_input("pH", min_value=6.8, max_value=7.8, value=7.34, step=0.01) input_values['pao2'] = st.number_input("PaO2", min_value=40, max_value=300, value=72) input_values['pco2'] = st.number_input("PCO2", min_value=20, max_value=100, value=58) input_values['fio2'] = st.number_input("FiO2", min_value=0.21, max_value=1.0, value=0.88, step=0.01) input_values['gcs'] = st.slider("GCS Score", 3, 15, 5) input_values['comorbidity_index'] = st.slider("Comorbidity Index", 0, 10, 1) input_values['admission_source'] = st.selectbox("Admission Source", [0, 1, 2, 3], index=1, format_func=lambda x: ["Emergency", "OPD", "Transfer", "Other"][x]) with col3: input_values['elective_surgery'] = st.selectbox("Elective Surgery", [0, 1], index=1, format_func=lambda x: "No" if x == 0 else "Yes") input_values['num_medications'] = st.number_input("Number of Medications", min_value=0, value=18) input_values['charlson_comorbidity_index'] = st.slider("Charlson Comorbidity Index", 0, 15, 1) input_values['ews_score'] = st.slider("EWS Score", 0, 20, 7) input_values['severity_score'] = st.slider("Severity Score", 0, 10, 4) input_values['bed_occupancy_rate'] = st.slider("Bed Occupancy Rate (%)", 50, 100, int(68.67)) input_values['staff_to_patient_ratio'] = st.slider("Staff to Patient Ratio", 0.1, 2.0, 0.99, step=0.1) input_values['past_icu_admissions'] = st.number_input("Past ICU Admissions", min_value=0, value=2) input_values['previous_surgery'] = st.selectbox("Previous Surgery", [0, 1], index=1, format_func=lambda x: "No" if x == 0 else "Yes") input_values['high_risk_treatment'] = st.selectbox("High Risk Treatment", [0, 1], index=1, format_func=lambda x: "No" if x == 0 else "Yes") input_values['discharge_support'] = st.selectbox("Discharge Support", [0, 1], index=0, format_func=lambda x: "No" if x == 0 else "Yes") if st.button("Predict Outcome"): # Define input columns (must match your model's expected input features) input_columns = [ 'age', 'gender', 'previous_hospitalizations', 'heart_rate', 'respiratory_rate', 'blood_pressure_sys', 'blood_pressure_dia', 'temperature', 'wbc_count', 'creatinine', 'bilirubin', 'glucose', 'bun', 'pH', 'pao2', 'pco2', 'fio2', 'gcs', 'comorbidity_index', 'admission_source', 'elective_surgery', 'num_medications', 'charlson_comorbidity_index', 'ews_score', 'severity_score', 'bed_occupancy_rate', 'staff_to_patient_ratio', 'past_icu_admissions', 'previous_surgery', 'high_risk_treatment', 'discharge_support' ] # Create a sample input for prediction (using a template from your dataset) if len(df) > 0: sample_input = pd.DataFrame([{col: 0 for col in input_columns}]) # Update with user inputs for feature, value in input_values.items(): if feature in sample_input.columns: sample_input[feature] = value # Make prediction try: prediction = model.predict(sample_input)[0] prediction_proba = model.predict_proba(sample_input)[0] # Display prediction st.subheader("Prediction Result") st.write(f"Predicted Class: {prediction} - {class_descriptions.get(prediction, 'Unknown')}") # Display probability for each class st.write("Prediction Probabilities:") proba_df = pd.DataFrame({ 'Class': [class_descriptions.get(i, f"Class {i}") for i in range(len(prediction_proba))], 'Probability': prediction_proba }) fig = px.bar(proba_df, x='Class', y='Probability', title='Prediction Probabilities') st.plotly_chart(fig) except Exception as e: st.error(f"Error making prediction: {e}") else: st.error("Dataset is empty, cannot create input template.") else: st.error("Failed to load model. Please check if 'xgboost_patient_model.json' exists in the current directory.") # NEW SECTION 1: Batch Prediction elif option == 'Batch Prediction': st.title("🏥 Batch Patient Outcome Prediction") st.write("Upload a CSV file with patient data to predict outcomes for multiple patients at once.") uploaded_file = st.file_uploader("📂 Upload CSV file", type=["csv"]) if uploaded_file is not None: batch_df = pd.read_csv(uploaded_file) batch_df = batch_df.dropna().reset_index(drop=True) st.write("## Preview of Uploaded Data") st.dataframe(batch_df.head(), use_container_width=True) model = load_model() actual_target, predicted_classes, predicted_outcomes, actual_outcomes = predict_outcome(model, batch_df.copy()) if predicted_classes is not None: st.write("## 🏥 Prediction Results") result_df = pd.DataFrame({ "Patient ID": range(1, len(predicted_classes) + 1), "Actual Class": actual_target if actual_target is not None else ["N/A"] * len(predicted_classes), "Predicted Class": predicted_classes, "Predicted Outcome": predicted_outcomes }) st.dataframe(result_df, use_container_width=True) # Add visualization of batch prediction results st.write("## Prediction Distribution") results_count = pd.Series(predicted_outcomes).value_counts().reset_index() results_count.columns = ['Predicted Outcome', 'Count'] fig = px.pie(results_count, values='Count', names='Predicted Outcome', title='Distribution of Predicted Outcomes') st.plotly_chart(fig) # Offer download of results csv_results = result_df.to_csv(index=False).encode('utf-8') st.download_button( label="Download Prediction Results", data=csv_results, file_name="patient_predictions.csv", mime="text/csv" ) # NEW SECTION 2: Model Performance elif option == 'Model Performance': st.title("📊 Model Performance Analysis") # Check if data exists and contains target variable if len(df) > 0 and 'target' in df.columns: st.write("Analyze the model's performance on the dataset.") # Split data into features and target X = df.drop(columns=["target"]) # Features y = df["target"] # Target # Split data for testing X_train, X_test, y_train, y_test = train_test_split( X, y, test_size=0.2, random_state=42, stratify=y ) # Load the model model = load_model() if model is not None: # Make predictions try: y_pred = model.predict(X_test) y_prob = model.predict_proba(X_test) # Calculate metrics accuracy = accuracy_score(y_test, y_pred) conf_matrix = confusion_matrix(y_test, y_pred) class_report = classification_report(y_test, y_pred, output_dict=True) # Display metrics col1, col2 = st.columns(2) with col1: st.metric("Model Accuracy", f"{accuracy:.2%}") # Plot confusion matrix st.write("### Confusion Matrix") fig, ax = plt.subplots(figsize=(10, 8)) sns.heatmap(conf_matrix, annot=True, fmt='d', cmap='Blues', ax=ax) ax.set_xlabel('Predicted Labels') ax.set_ylabel('True Labels') st.pyplot(fig) with col2: # Plot classification report st.write("### Classification Report") report_df = pd.DataFrame(class_report).transpose() st.dataframe(report_df.style.format({ 'precision': '{:.2f}', 'recall': '{:.2f}', 'f1-score': '{:.2f}', 'support': '{:.0f}' })) except Exception as e: st.error(f"Error performing analysis: {e}") else: st.error("Model could not be loaded. Please check if the model file exists.") else: st.error("Cannot perform model analysis. Dataset is empty or missing target variable.") st.sidebar.write("Forecasting discharge outcomes for critically ILL patients using machine learning")