مقاله انگلیسی رایگان در مورد الگوریتم جستجو گرانشی دوگانه بازگشتی – الزویر 2019

1. Introduction

2. Proposed cancer classification

3. Experimental Setup and Results

4. Conclusion

Acknowledgments

References

Nowadays, in clinical medicine diagnosticians usually use DNA microarray datasets for diagnosis and classification of cancer. However, DNA microarray datasets typically have very large number of genes and less number of samples, therefore, before diagnosis and classification of cancer it is quite requisite to select most relevant genes. In this paper, we have developed a two phase classification model in which most relevant genes are selected by integrating ReliefF with Recursive Binary Gravitational Search Algorithm (RBGSA) in the help of a classifier of Multinomial Naive Bayes. The RBGSA recursively transforms a very raw gene space to an optimized one at each iteration while not degrading the accuracy. We evaluate our model by comparing it with 6 other known methods on 6 different microarray datasets of cancer. Comparison results show that our model gets substantial improvements in accuracy over other methods.

Introduction

Filter methods include univariate filters and multivariate filters. Univariate filters search and evaluate each gene separately by surveying its inherent natures with regard to discriminate class, thus leading to unreliable outcomes because of not considering gene interactions. While multivariate filters search and evaluate the subset of genes through surveying their inherent natures with regard to different classes, which can promise better results than univariate filters in identifying the most relevant genes in microarray data. Relief is one of the multivariate filter approaches [3,4] based property ranking scheme. Kononenko later developed an improved method called ReliefF based on Relief [5]. In many classification tasks, Relief and ReliefF are usually used as pre-processing approaches for feature selection prior to the model learning. These types of methods not only are effective but also are able to accurately assess the importance of properties [6].