A Novel Hybrid Approach to Contraceptive Demand Forecasting: Integrating Point Predictions with Probabilistic Distributions

TL;DR

This paper tackles the gap in contraceptive demand forecasting by integrating expert judgment with probabilistic forecasting in the context of frontline health supply chains. It introduces Constrained Quantile Regression Averaging (CQRA) to combine point forecasts from domain experts with probabilistic outputs from multiple models, producing coherent forecast distributions and actionable uncertainty estimates. Through a comprehensive experimental setup on Côte d'Ivoire LMIS data, the study demonstrates that ML-based methods generally outperform traditional time-series approaches, while the proposed hybrid Weighted Averaging method delivers strong point and probabilistic performance with practical runtime characteristics. The work offers a practical framework and guidelines for FPSC decision-making and is readily adaptable to other humanitarian contexts facing data constraints and demand uncertainty.

Abstract

Accurate demand forecasting is vital for ensuring reliable access to contraceptive products, supporting key processes like procurement, inventory, and distribution. However, forecasting contraceptive demand in developing countries presents challenges, including incomplete data, poor data quality, and the need to account for multiple geographical and product factors. Current methods often rely on simple forecasting techniques, which fail to capture demand uncertainties arising from these factors, warranting expert involvement. Our study aims to improve contraceptive demand forecasting by combining probabilistic forecasting methods with expert knowledge. We developed a hybrid model that combines point forecasts from domain-specific model with probabilistic distributions from statistical and machine learning approaches, enabling human input to fine-tune and enhance the system-generated forecasts. This approach helps address the uncertainties in demand and is particularly useful in resource-limited settings. We evaluate different forecasting methods, including time series, Bayesian, machine learning, and foundational time series methods alongside our new hybrid approach. By comparing these methods, we provide insights into their strengths, weaknesses, and computational requirements. Our research fills a gap in forecasting contraceptive demand and offers a practical framework that combines algorithmic and human expertise. Our proposed model can also be generalized to other humanitarian contexts with similar data patterns.

Figures (10)

• Figure 1: Contraceptive stock distribution in Côte d'Ivoire by healthcare site location. The size of the circles represents the quantity of stock distributed.
• Figure 2: Time series of contraceptive product stock distributed (Jan 2016 -- Dec 2019) at various levels. The x-axis represents the month, while the y-axis indicates the number of units distributed. The panels display data from the entire country (top panel), with breakdowns by region, district, site, and product code. The bottom panel shows the number of units distributed in selected sites for specific products. To ensure clarity and prevent overplotting, only five time series are displayed for each aggregate level. These series were selected randomly and are characteristic of the patterns encountered at the respective aggregation levels.
• Figure 3: Time series of contraceptive product stock distributed in selected sites for specific products(Jan 2016 -- Dec 2019). To ensure clarity and prevent overplotting, only five of the products are displayed. These series were selected randomly and represent characteristic patterns at this level.
• Figure 4: Trend strength and seasonality in the time series of stock distribution. Each point in the scatter plot represents one of the 1,360 time series analysed, with trend and seasonality strengths measured on a scale from 0 to 1 (0 indicating weak and 1 indicating strong).
• Figure 5: Average ranks of forecasting methods with 95% confidence intervals based on the Nemenyi test for MASE values. Lower ranks indicate better forecast performance.