Study on sensor fault tolerance with multi-classifiers of SVM fusion

In some application fields of industry and agriculture, sensor is the main tool for getting information. When some sensor is wrong, the performance of a single decision system will get down sharply, even all the system will collapse. In order to improve the fault tolerance, the multi-classifiers fusion method is introduced into this field. Firstly, 3 different reductions were gotten through rough sets. Then with fuzzy output SVM method, 3 classifiers were trained. The membership to each class of every testing sample was gotten through the above 3 classifiers. With the average value fusion method, the final class was given. In the experiment of 6 UCI standard data sets, it shows when some sensor is open or short, the performance of the single classifier will get down, while with the fusion method, the accuracy is comparable to the original accuracy.