Interface 2003

Discovery of Battle States Knowledge from Muti-Dimensional Time Series Data
T.W. Liao, (Louisianna State University),,
B. Bodt, (U.S. Army Research Laboratory),,
J. Forester, (U.S. Army Research Laboratory),,
C. Hansen, (U.S. Army Research Laboratory),,
E. Heilman, (U.S. Army Research Laboratory),,
C. Kaste, (U.S. Army Research Laboratory),, and
J. O'May, (U.S. Army Research Laboratory),


This paper presents a suite of knowledge discovery techniques developed for processing multi-dimensional battlespace time series data in order to capture the knowledge of determining battle states in the form of fuzzy rules. This discovered knowledge is subsequently applied to provide the battle staff (or the nerve center of the Future Combat System) with current battle state information and to predict future battle outcomes. The major techniques employed include interpolation for preprocessing data, clustering to establish the concept of battle states, and genetic-fuzzy modeling for time series forecasting and classification. The clustering results are first checked by human experts and then used as the input data for the discovery of battle state determination. Two knowledge discovery methods have been implemented: WM-based genetic-fuzzy and LIAO-based genetic fuzzy. Represented in the form of fuzzy rules, the discovered knowledge is very easy to comprehend if the number of fuzzy terms is not too high. The predicted data are fed into the knowledge of determining battle states to predict future battle states. This is easily achieved by a fuzzy reasoning method such as max-min.

Data are needed to test the effectiveness of this tool. To this end, we created a battle scenario using the OneSAF combat simulation. In the full paper, more details will be given to describe the experiments and the Killer-Victim Scoreboard (KVS) data collection method. The discovered knowledge and their use will also be presented as well.

Take me back to the main conference page.