My project is to develop models for understanding and predicting decisions taken by pastoral (animal herders) communities in response to changes in Nature and their environment. The subjects of the study are East African pastoralists who have biannual dry seasons and need to migrate due to it. A survey was collected by the USAID Global Livestock Collaborative Research Support program (GL CRSP) which will be used as data to model this system.
The objective is to use data on herd movement choices to estimate the parameters of a spatially explicit dynamic model. This model estimates the household herd allocation over space and time when there is herd split or migration caused by some natural phenomena or search for water bodies. We need a model that derives the preferences of real pastoral populations.
There have been previous attempts to study the system using econometric techniques. But given the complex nature of the problem, we plan to approach the problem using advanced Machine Learning techniques like Reinforcement Learning and Inverse Reinforcement Learning. I aim to apply such techniques to develop a decision making model. The data embodies a policy and is indicative of the choices made by pastoral communities. This policy will be used to generate a reward function (corollary to utility function in economics). A policy coupled with reward function would be sufficient to describe the preferences of the pastoralists.