RCGA-S/RCGA-SP Methods to Minimize the Delta Test for Regression Tasks

Abstract

Abstract. Frequently, the number of input variables (features) involved in a problem becomes too large to be easily handled by conventional machine-learning models. This paper introduces a combined strategy that uses a real-coded genetic algorithm to find the optimal scaling (RCGA-S) or scaling + projection (RCGA-SP) factors that minimize the Delta Test criterion for variable selection when being applied to the input variables. These two methods are evaluated on five different regression datasets and their results are compared. The results confirm the goodness of both methods although RCGA-SP performs clearly better than RCGA-S because it adds the possibility of projecting the input variables onto a lower dimensional space.

Approximate k-NN delta test minimization method using genetic algorithms: Application to time series

“…From previous analysis [37,58], the best crossover and mutation rates for a feature selection application using the GA are 0.85 and 0.1 respectively, and an elitism of 10% of the individuals was the best compromise.…”

“…This value appears to be a good compromise between performance and computational cost for GA-based search applied to similarly sized datasets [37,58]. After some preliminary tests, the number of generations was fixed to 200 to ensure convergence in all cases.…”

The delta test: The 1-NN estimator as a feature selection criterion

“…Experimental evidence comparing to alternative methods is also provided. Choosing an efficient search scheme for high-dimensional tasks is mostly left as a practical matter of implementation, and several papers specifically about optimising the Delta test have also been published in the literature [13], [14], [15], [16], [17], [18], [19].…”

Redes neuronales y preprocesado de variables para modelos y sensores en bioingeniería

“…This value appears to be a good compromise between performance and computational cost for GA-based search applied to similarly sized datasets [135,227]. After some preliminary tests, the number of generations was fixed to 200 to ensure convergence in all cases.…”