Research On Multi-objective Feature Selection Based On Improved Differential Evolution Algorithm

In the information age,the rapid growth of data over time has brought new challenges to data mining and machine learning tasks.In real tasks,we often encounter the "curse of dimensionality" problem caused by too many attributes in data.Therefore,in order to extract useful information from large data repositories and make further decisions and predictions within a limited time,high-dimensional data must be dimensionally reduced.In addition,the increase in the size of the dataset has also brought about the problem of feature redundancy,and the poor correlation between certain features has become an obstacle to effectively processing the dataset.Therefore,removing redundant and poorly correlated features is crucial for reducing the difficulty of learning tasks.Feature Selection(FS)can effectively reduce the dimensionality of data,remove redundant features,especially in solving classification problems on high-dimensional datasets.In practical applications,the optimization objectives of feature selection may be multiple,such as classification accuracy and the size of the dataset.However,in multi-objective feature selection,reducing classification error rates and minimizing the number of selected features are conflicting.In order to balance these two goals and improve the time complexity of the algorithm,the following research was mainly conducted:(1)By studying and comparing different feature selection search strategies,a differential evolution algorithm with random search strategy and high efficiency will be used as the basis for feature selection.However,the mutation operator and crossover probability of differential evolution algorithms are random values selected within a specific interval,and existing improved algorithms for both are more complex and may lead to premature convergence issues.Therefore,this article proposes an adaptive differential evolution algorithm and a new parameter adaptive strategy,which updates the mutation operator and crossover probability in a timely manner as the number of iterations changes.It is applied to multi-objective feature selection and named as the multi-objective feature selection algorithm based on adaptive differential evolution.(2)In order to further improve the global optimization ability of multi-objective feature selection algorithms,an efficient non-dominated sorting strategy was introduced in the selection stage of the adaptive differential evolution algorithm to enhance the global competitiveness of individuals.A multi-objective feature selection algorithm based on adaptive differential evolution and efficient non dominated sorting was proposed.The algorithm sorts the population based on objects values and non-dominated relationships,forming Pareto frontiers,multiple optimal feature subsets are formed for selection.The experimental results show that the multi-objective feature selection algorithm based on adaptive differential evolution proposed in this paper enriches solutions diversity,and the multi-objective feature selection algorithm based on adaptive differential evolution and efficient non-dominated sorting has good global optimization ability.Compared with some traditional multi-objective feature selection algorithms,it can reduce the time complexity of the algorithm.