Comparison of the Kalman filter and exponential smoothing techniques of forecasting United States Marine Corps losses in the Republic of Vietnam

Abstract

This paper investigates the application of the Kalman Filter and the General Exponential Smoothing techniques of forecasting. Both methods are derived and the similarities and differences between them are discussed. The two techniques are then applied to the practical problem of predicting weekly losses suffered by the U. S. Marine Corps units in the I Corps Tactical Zone in the Republic of Vietnam. The mean absolute error of the prediction is used as the criterion for choosing the better of the two methods. Results are given for both techniques as well as for the method of linear regression. In general the Kalman Filter provides the smallest mean absolute error for the three mathematical models; linear, growing sine with harmonics and frequency of sixteen, thirty-two, and fifty-two weeks, and a constant model.