The role of recall periods when predicting food insecurity: A machine learning application in Nigeria

Defining and measuring food insecurity at the household level is critical for policymakers, aid agencies, and international organizations. Food insecurity indicators, such as the Food Consumption Score, are based on a given recall period, usually 7-days. Still, using other indicators or methodologies makes surveys use different recall periods (e.g., 30-days or 12 months). This study uses machine learning methods and four waves of the Nigeria LSMS-ISA datasets to assess the implications of using different recall periods when predicting food insecurity measures. In addition to machine learning methods, the novelty of this study is the use of big data and relevant weather data to predict the food insecurity status of Nigerian farming families. Our results show that experience-based food insecurity indicators, measured using a 7-days recall period, have a high predictability accuracy (78%–90%). More importantly, we find that predictors computed using a 7-day recall period can detect about seven out of ten households considered food-insecure by indicators measured using a recall period of 30-days.