Enhancing classification accuracy using elite breeding QPSO on gene dataset
Himanshu Agarwal, Poonam Chaudhari · 2017
This paper focuses on feature set selection on microarray gene expression for cancer classification. The gene expression data consists of high number of genes for several samples. The imbalance between the number of genes and the number of samples makes it crucial to introduce algorithms for precise feature selection. We propose to work with Quantum Particle Swarm Optimization with elitist breeding (EBQPSO) on gene datasets. To the best of our understanding, the exploration of the elitist is not taken into account for genetic datasets. As the algorithm works with the elitist of the swarm, the problem of local minima is solved. A transposon operator is used to form new set of individuals' with the help of personal best of each particle and the global best of the swarm. As a contribution, we have introduced a random variable in the standard converse formula in order to compute the outliers. This addition helps us handle boundary violation condition which leads to faster convergence rate as restoration is not required. This modification in the algorithm takes care of Overlapping condition, in turn reduces the time complexity. The results show that EBQPSO with remodeled transposon operator performs better than the previous variants of PSO.