Topic 15 – Healthcare Call Data Analysis During Emergency Times (Kaggle)

Dataset

• Source: Healthcare Call Data Analysis During Emergency Times – Kaggle
• Link: https://www.kaggle.com/datasets/shuvokumarbasak2030/healthcare-call-data-analysis-duringemergencytimes/data

Download Instructions

1. Open the dataset link above.
2. Log in to Kaggle.
3. Click "Download".
4. Extract to data/healthcare_calls/.
5. Use the file daily_and_month_call_report.csv.

Data Loading

import pandas as pd

df = pd.read_csv("data/healthcare_calls/daily_and_month_call_report.csv")

# Create datetime index from Year and Month columns
df['Date'] = pd.to_datetime(df['Year'].astype(str) + '-' + df['Month'] + '-01')
df = df.set_index('Date').sort_index()

# Select a time series to analyze (e.g., Total Number of Calls)
ts = df['Total Number of Calls']

Implementation Steps

1. Data Exploration

• Load the dataset and inspect structure
• Understand the multiple time series available (calls, consultations, health info, ambulance info, complaints, service inquiries)
• Select one or more time series for analysis (e.g., Total Number of Calls)
• Visualize the time series over time
• Check for missing values and handle them appropriately
• Examine basic statistics (mean, std, min, max)
• Identify patterns, especially during emergency periods (e.g., 2021-2022)

2. Exploratory Data Analysis (EDA)

• Plot time series with trend lines
• Identify emergency periods and their impact on call volumes
• Perform time series decomposition (trend, seasonality, residual)
• Calculate and visualize ACF/PACF plots
• Identify monthly and yearly seasonality patterns
• Analyze outliers or unusual patterns (especially during emergency times)
• Compare call volumes across different emergency periods

3. Stationarity Analysis

• Test for stationarity using ADF test
• Apply differencing if needed (1st or 2nd order)
• Consider seasonal differencing if strong monthly/yearly patterns exist
• Consider segmenting data by emergency vs normal periods
• Apply log transformation if variance is non-constant

4. Model Building

Traditional Methods:

• ARIMA Models:
• Use ACF/PACF to identify initial (p, d, q) parameters
• Try multiple ARIMA configurations
• Use AIC/BIC for model selection
• SARIMA Models:
• Identify seasonal patterns (monthly: 12 months, yearly patterns)
• Fit SARIMA models with seasonal components (e.g., SARIMA(p, d, q)(P, D, Q)12)
• Compare different seasonal configurations
• Consider models that account for emergency period effects

Machine Learning Methods:

• Feature Engineering:
• Create lag features (previous month, year)
• Add time-based features (month, year, quarter, is_emergency_period)
• Rolling statistics (3-month, 12-month averages)
• Create dummy variables for emergency periods
• Models:
• Random Forest or XGBoost for regression
• Linear regression with features
• Compare with ARIMA/SARIMA baseline

Deep Learning Methods (Optional):

• LSTM/GRU:
• Prepare sequences for RNN models
• Train LSTM or GRU networks
• Tune hyperparameters
• Simple CNN:
• Use 1D convolutions for time series
• Compare with RNN approaches

5. Model Evaluation

• Split data temporally (e.g., last 20-30% as test set)
• Use walk-forward validation for time series
• Calculate metrics: MAE, RMSE, MAPE
• Compare traditional vs ML/DL approaches
• Visualize actual vs predicted values
• Analyze forecast errors and residuals

6. Forecasting

• Generate future forecasts (e.g., next 6-12 months)
• Include prediction intervals/confidence bands
• Visualize forecasts with historical data
• Interpret results in healthcare emergency context
• Consider scenario analysis for future emergency periods
• Consider scenario analysis for future emergency periods

Expected Deliverables

1. EDA Report:

• Time series plots showing patterns
• Decomposition plots (trend, seasonal, residual)
• ACF/PACF plots
• Stationarity test results
• Monthly/yearly pattern analysis
• Comparison of emergency vs normal periods

1. Model Results:

• Best traditional model (ARIMA/SARIMA) with parameters
• Best ML model with feature importance
• Best DL model (if implemented) with architecture
• Performance comparison table
• Forecast plots with confidence intervals
• Analysis of model performance during different periods

1. Code:

• Complete Python notebook
• Functions for data preprocessing, feature engineering, modeling
• Visualization utilities
• Model comparison scripts

Tips

• The dataset contains monthly data, not daily - adjust your analysis accordingly
• Pay special attention to emergency periods (e.g., 2021-2022) which show significant spikes
• Consider creating separate models for emergency vs normal periods
• Monthly seasonality (12-month patterns) is likely important
• Feature engineering should include emergency period indicators
• Use appropriate train/validation/test splits (temporal, not random)
• Start with simple models (ARIMA) before moving to complex ML/DL
• Feature engineering is crucial for ML models - create meaningful time-based features
• Compare multiple approaches and justify your choice
• Document any data preprocessing steps clearly
• Consider the healthcare context when interpreting results