Reliable Attribute Selection Based on Random Forest (RASER)

Multi‐objectives optimisation of features selection for the classification of thyroid nodules in ultrasound images

“…To overcome this problem, we used an optimisation method to retain only the most significant features [33]. This method is based on multi‐objective optimisation (MbPSO) and in particular on random forest (RF) [34] and has three stages: measuring the relevance of the features, measuring the redundancy and optimisation.…”

“…An RF is a set of binary decision trees which, in general, are more efficient than simple decision trees but have the disadvantage of being more difficult to interpret. Each feature is represented in the RF by a node of a decision tree [34].…”

“…Optimisation : To optimise the subset of extracted features, we construct a complete and weighted graph G from the dissimilarity matrix, where the nodes represent the features and the edges (weight) represent the similarity between the features. To do this, We apply a multi‐objective binary PSO [34] G, to find an optimal sub graph which contains only the most significant features. The extracted features have been optimised using this optimisation algorithm.…”