What is the income of farming households and does it afford a decent living standard? This is a key question to answer in the design, monitoring, and evaluation of interventions seeking to improve livelihoods of farmers at the starting points of global supply chains such as cocoa and coffee. The answer requires primary data collection using structured household surveys. Large-scale household surveys, however, are elaborate, costly, and time-consuming, which can make them restrictive in some cases.
Data science can help answer this question. KIT’s Monitoring and Evaluation for Social Impact team has worked on prototyping a machine learning model to estimate the net income of cocoa and coffee growing households in Côte d’Ivoire, Ghana, and Ethiopia. The team mined KIT’s data archives on crop production and income. Using these data and our experience with the drivers of farmer’s income to select and test variables for predictive model. We trained a machine learning model on farm location, household size, and crop production—simple variables known to influence net household income. In its prototype state, the model has shown great accuracy (with a root-mean-square error < $10USD in the testing dataset) and, therefore, potential for application.
The goal of our data science services for living income work is to make monitoring and evaluation of household income cost-effective and scalable. With our sound method for monitoring and evaluation and machine learning model, we offer an approach that improves its performance with use and frees up resources to design pathways to close the income gap using, for example, segmented approaches.
Are you interested in using machine learning in your living income project? Contact our advisors.