Mathematics and Statistics Vol. 11(6), pp. 895 - 909
DOI: 10.13189/ms.2023.110604
Reprint (PDF) (1507Kb)


An Optimal Approach to Identify the Importance of Variables in Machine Learning Using Cuckoo Search Algorithm


Asep Rusyana 1,2, Aji Hamim Wigena 1, I Made Sumertajaya 1, Bagus Sartono 1,*
1 Department of Statistics, Faculty of Mathematics and Natural Sciences, IPB University, Indonesia
2 Department of Statistics, Faculty of Mathematics and Natural Sciences, Universitas Syiah Kuala, Indonesia

ABSTRACT

Different machine learning algorithms may produce different orders of the variable importance measures even though they use an identical dataset. The measures raise the difficulty of concluding which predictor variables are the most important. Therefore, there is a requirement to unify those scores into a single order so that the analyst can withdraw a conclusive decision more easily. This research applied the Cuckoo Search algorithm approach to obtain the unification of those orders into a single one. A simulation study was conducted to justify that the approach could work well in several circumstances of data. We implemented the algorithm to identify the importance of the variables where the correlations among them are low, moderate, and high. The result of the paper shows that the proposed variable importance measure is the best if it is applied to predictors independent of each other. Generally, it is more accurate than variable importance measures of machine learning. The algorithm was also applied to identify the proposed important variable measure for recognizing food insecurity in households in Indonesia. The proposed variable importance has good accuracy. The accuracy is higher if the number of variables is greater than ten.

KEYWORDS
Machine Learning, Variable Importance Measurement, Simulation Data, Cuckoo Search

Cite This Paper in IEEE or APA Citation Styles
(a). IEEE Format:
[1] Asep Rusyana , Aji Hamim Wigena , I Made Sumertajaya , Bagus Sartono , "An Optimal Approach to Identify the Importance of Variables in Machine Learning Using Cuckoo Search Algorithm," Mathematics and Statistics, Vol. 11, No. 6, pp. 895 - 909, 2023. DOI: 10.13189/ms.2023.110604.

(b). APA Format:
Asep Rusyana , Aji Hamim Wigena , I Made Sumertajaya , Bagus Sartono (2023). An Optimal Approach to Identify the Importance of Variables in Machine Learning Using Cuckoo Search Algorithm. Mathematics and Statistics, 11(6), 895 - 909. DOI: 10.13189/ms.2023.110604.