Probabilistic reasoning in intelligent systems: networks of plausible inference
Probabilistic reasoning in intelligent systems: networks of plausible inference
The sensitivity of belief networks to imprecise probabilities: an experimental investigation
Artificial Intelligence - Special volume on empirical methods
On the Optimality of the Simple Bayesian Classifier under Zero-One Loss
Machine Learning - Special issue on learning with probabilistic representations
Why is diagnosis using belief networks insensitive to imprecision in probabilities?
UAI'96 Proceedings of the Twelfth international conference on Uncertainty in artificial intelligence
| Hi-index | 0.00 |
Most real-world applications of diagnosis involve continuous-valued attributes, which are normally discretized before the existing classification algorithms are applied. The discretization may be based on data or on human expertise. In cellular networks the number of classified examples is very limited. Thus, the diagnosis experts should specify the boundaries of the intervals for each discretized symptom. The large number of values makes it difficult to specify precise parameters. Even if boundaries are obtained from classified examples, due to the limited number of cases, the obtained values are not very accurate. In this paper two techniques to improve the performance of diagnosis systems based on Bayesian Networks are compared. Some empirical results are presented for diagnosis in a GSM network. The first method, Smooth Bayesian Networks, is shown to be more robust to imprecise setting of boundaries. The second method, Multiple Uniform Intervals, is superior if accurately defined boundaries are available.