Importance analysis of decision making factors based on fuzzy decision trees

Classification is one important and commonly used Machine Learning technique. The classification effectiveness can be influenced by different factors and initial data used for the classifier induction and in the process of classification, including. The quality of measurements of the input attributes of new samples for the classification can affect the accuracy and reliability of the classification result. At the same time, the degree of influence of different attributes is not the same. There are some attributes that are most important for the classification because have a greater influence on the classification result than others. A new method for the determination of the most important attributes is proposed. This method is developed based on the approach of Importance Analysis, which is widely used in reliability engineering. The proposed method investigates the sensitivity of the classification from the input attributes and indicates the attributes for which changes have the most impact on the classification result. The attribute’s importance is evaluated by the special index which is known as structural importance in reliability engineering. This method application is illustrated in the paper by Fuzzy Decision Tree, but it can be used for any other classifiers induction. The use of a fuzzy classifier allows for taking into consideration of possible uncertainty of initial data.