Predicting battle outcomes with classification trees

• Main Author: Coban, Muzaffer.
• Other Authors: Lucas, Thomas W.
• Format: Others
• Published: Monterey, California. Naval Postgraduate School 2012
• Online Access: http://hdl.handle.net/10945/1342

Historical combat data analysis is a way of understanding the factors affecting battle outcomes. Current studies mostly prefer simulations that are based on mathematical abstractions of battles. However, these abstractions emphasize objective variables, such as force ratio. Models have very limited abilities of modeling important intangible factors like morale, leadership, and luck. Historical combat analysis provides a way to understand battles with the data taken from the actual battlefield. The models built by using classification trees reveal that the objective variables alone cannot explain the outcome of battles. Relative factors, such as leadership, have deep impacts on success. This result suggests that combat simulations will have a difficult time predicting combat outcomes unless we can better account for these intangible factors. Historical combat analysis helps us comprehend these factors. The classification model predictions on test sets reveal correct classification rates as high as 79 percent. Considering the variability in the data set this outcome is satisfying. Classification models also reveal that the factors affecting outcome of battles have changed throughout history. The leadership advantage played an important role for hundreds of years. However, in the 20th century, air sorties, tanks, and intelligence showed a higher importance. === Turkish Army author