How agent based models can be utilized to explore and exploit non-linearity and intangibles inherent in guerrilla warfare

Abstract

Since the end of WWII, a host of groups and states have pursued their interests in the Low Intensity Conflict (LIC) environment. One of the characteristics of LIC is that it is executed mostly by the rules of asymmetric war or guerrilla warfare. This thesis utilizes the recently developed agent-based model Map Aware Non-Uniform Automata (MANA) to explore nonlinearity and intangibles inherent in guerrilla warfare. An infiltration scenario is developed based on the author's experiences fighting guerrillas in the mountains of Southeast Turkey. To simultaneously investigate the effects of as many as 22 input variables, recently developed Near Orthogonal Latin Hypercube Designs and Fractional Factorial Designs are used. Utilizing a personal computer and the computational capabilities of supercomputers run by Mitre for the Marine Corps Combat Development Center (MCCDC), approximately 200,000 MANA runs were completed. Several statistical models are developed and compared using a variety of diverse statistical techniques, including Cluster Analysis, Neural Networks, Regression Trees, Linear Regression, and Bayesian Networks. The results of the analysis suggest that the outcome of an infiltration scenario is heavily dependent on the Red agent parameters. The analysis also reveals the Red Stealth parameter as the most important factor in predicting the MOEs.