Datasets

Workshop Data Resources

Supply Chain Dataset

The primary dataset for hands-on exercises simulates pharmaceutical supply chain data across Ethiopia’s health facilities.

Description

Variable Type Description
facility_id string Unique facility identifier (F001-F020)
facility_name string Facility name
facility_type categorical Hospital, Health Center, or Clinic
region string Geographic region (5 regions)
population_served numeric Population catchment area
month numeric Month of observation (1-12)
year numeric Year of observation
previous_demand numeric Demand in the previous month
distance_to_warehouse numeric Distance to regional warehouse (km)
season categorical Dry or Rainy season
stockout_last_month binary 1 if stockout occurred last month
avg_delivery_days numeric Average delivery time (days)
storage_capacity numeric Maximum storage capacity (units)
actual_demand numeric Actual demand for the month (target variable)
shortage_frequency numeric Historical shortage count for this facility (0-5)
shortage_occurred binary 1 if shortage occurred this month
shortage_duration_days numeric Duration of shortage in days (0 if none)
shortage_duration_category categorical None, Short (1-7 days), Medium (8-21 days), Long (22+ days)
shortage_cause categorical None, Manufacturing, Logistics, Demand_Surge, Regulatory, Raw_Materials

Regional Coverage

Region Facilities Facility Types
Addis Ababa 4 2 Hospitals, 2 Health Centers
Oromia 4 2 Hospitals, 1 Health Center, 1 Clinic
Amhara 4 2 Hospitals, 1 Health Center, 1 Clinic
SNNP 4 2 Hospitals, 1 Health Center, 1 Clinic
Tigray 4 2 Hospitals, 1 Health Center, 1 Clinic

Loading the Data

In Google Colab:

import pandas as pd

# From GitHub (recommended)
url = "https://raw.githubusercontent.com/sysylvia/ethiopia-ds-workshop-2026/main/data/supply-chain-sample.csv"
df = pd.read_csv(url)

print(f"Loaded {len(df)} records from {df['facility_id'].nunique()} facilities")

Data Characteristics

  • Time range: January - December 2024 (12 months)
  • Facilities: 20 facilities across 5 regions
  • Observations: 240 rows (20 facilities × 12 months)
  • Target variable: actual_demand (units of medication)

Key Patterns in the Data

  1. Seasonal effects: Demand decreases during rainy season (months 5-8)
  2. Facility type effects: Hospitals have higher demand than clinics
  3. Regional variation: Distance to warehouse affects delivery times
  4. Temporal autocorrelation: Previous demand predicts current demand
  5. Shortage patterns: ~20% of facility-months experience shortages, with most being short duration

Shortage Variables (for Notebook 05)

These columns support the Roe et al. (2025) replication exercise:

Variable Distribution Notes
shortage_frequency Mean ~1.7 Higher for clinics and distant facilities
shortage_occurred ~20% positive Correlated with rainy season, stockout history
shortage_duration_category 53% Short, 32% Medium, 15% Long Among shortage events
shortage_cause Logistics most common Varies by distance and season

Sample Data Preview

df[['facility_name', 'region', 'month', 'previous_demand', 'actual_demand']].head(10)
facility_name region month previous_demand actual_demand
Yekatit 12 Hospital Addis Ababa 1 180 195
Yekatit 12 Hospital Addis Ababa 2 195 188
Yekatit 12 Hospital Addis Ababa 3 188 201
… … … … …

Data Quality Notes

  • This is synthetic data created for educational purposes
  • Patterns are inspired by real-world Ethiopian supply chain dynamics
  • The data includes realistic features like:
    • Seasonal demand patterns
    • Facility-type effects
    • Stockout events
    • Regional infrastructure differences
  • No actual health data or PHI is included

Relationship to ABM

The ML predictions from this dataset feed into the agent-based model:

  • Facility agents use expected_demand to determine order quantities
  • Warehouse agents manage regional stock based on aggregate forecasts
  • Uncertainty estimates from ML models inform safety stock levels

Additional Datasets

Additional datasets may be shared via Google Drive during the workshop:

  • ABM parameter files
  • Historical consumption patterns
  • Regional infrastructure characteristics