Research On Feature Selection Method Based On Forest Optimization Algorithm

With the development of science and technology,the amount of data processed by various industries is increasing exponentially every day.and the dimension of data is becoming larger and larger.Thus,there are a lot of redundant and irrelevant features in these data,which brings great challenges to machine learning,pattern recognition and data mining.Learning algorithms often encounter performance issues when dealing with high-dimensional data.Feature selection has been widely concerned as an effective way to solve this problem.In the development of feature selection methods,many effective optimization algorithms have been emerged,and Forest Optimization Algorithm(FOA)is one of them.The FOA is inspired by the evolution of natural forests.The algorithm simulates the forest evolution process.It is simple,easy to implement,fast convergence and high search efficiency.Although feature selection algorithm based on forest optimization algorithm(FSFOA)has achieved satisfactory results in solving feature selection problems.However.there are still some deficiencies.In this paper,based on the deficiencies proposed,we have improved the FSFOA.The primary research work of this paper is as follows:First,an improved FSFOA is proposed.This method introduces a contribution degree strategy embedded in the forest optimization algorithm.Its main purpose is to guide the forest optimization algorithm to search the optimal solution according to the correlation of the class and the redundancy of feature,which improves the search efficiency of the forest optimization algorithm in the feature selection problem.According to the relationship between the current tree and the current global optimal tree,a distance adaptive strategy is proposed.This strategy can effectively guide the algorithm to search the optimal tree quickly,which improves the search efficiency of the algorithm.In order to avoid the algorithm falling into the local optimal solution,the fitness function is improved.This function makes the forest optimization algorithm not only consider the performance of the entire feature subset when selecting features,but also consider the quality of each feature in the feature subset.We selected 10 data sets that are commonly used to verify the validity of the algorithm from the UCI data set.and compared experiments with some feature selection algorithms in recent years.Experimental results show that our proposed algorithm outperforms these feature selection methods.Second,feature selection of forest optimization algorithm based on a local search strategy is proposed.This method uses a local search strategy to guide the forest optimization algorithm,which selects as many high-quality features as possible and eliminates as many low-quality features as possible in the process of searching for an optimal solution.This strategy greatly ensures that each feature subset in the forest has a higher quality and improves the search efficiency of the algorithm.In order to initialize all the trees in the forest in an ideal position,the feature subset size determination mechanism is used.This mechanism can ensure that the number of features selected in each feature subset of the forest is relatively small during the initialization phase,which greatly improves the search efficiency of the algorithm.Finally,the experimental results prove the effectiveness of our proposed algorithm.