This work focus on situation prediction in data fusion systems.
A hypothesis evaluation algorithm based on artificial neural networks is introduced. It is evaluated and compared to an algorithm based on Bayesian networks which is commonly used. It is also compared to a simple "dummy" algorithm.
For the tests, a computer based model of the environment, including protected objects and enemy objects, is implemented. The model handles the navigation of the enemy objects and situational data is extracted from the environment and provided for the hypothesis evaluation algorithms.
It was the belief of the author that ANNs would be suitable for hypothesis evaluation if a suitable data representation of the environment were used. The representation requirements include pre processing of the situational data to eliminate the need for variable input size to the algorithm. This because ANNs poorly handles this; the whole network have to be retrained each time the amount of input data changes.
The results show that ANNs performed best of the three and hence seems to be suitable for hypothesis evaluation.