In view of the over-fitting problems caused by the high dimension of feature attributes of small sample sets and noise data

etc

during the construction of CART (classification and regression tree) classification tree

the rough set (RS) attribute reduction based on mutual information was introduced in the process of CART decision tree algorithm training. The similarity between information entropy and GINI coefficients to characterize the “purity” of the sample set was considered

attribute reduction on historical fault data was performed to reduce the attribute dimensions and optimize the training sets.On this basis

a classification decision tree was constructed and the output rules were visualized.Result showed that

when the improved CART decision tree algorithm was applied to the fault diagnosis of a certain type of aeroengine oil

the extracted rules were highly interpretable and can reduce the influence of redundant attributes and noise on decision.Compared with the common fault diagnosis algorithms

the diagnostic accuracy of the model was improved by about 20%

and AUC (area under curve) value was as high as 92%

enabling to effectively deal with the problem of small sample of high-dimensional discrete aeroengine faults.