a novel optimized feature selection technique for software projects effort estimation

a novel optimized feature selection technique for software projects effort estimation

recently, estimating the effort required for software development is of more interest. relying on some previously completed software project data is a common way to do effort estimation. many types of researches have been done to find out the most accurate estimation model. expert judgment, algorithmic and machine learning methods are the commonest models. doing feature selection with the aim of machine learning methods have great influence on the estimation accuracy. this study assays the use of an inclusive effort estimation framework to select an optimal features subset and also a set of optimization algorithms for each of the datasets and estimation methods.

simulation is carried out through a training and a testing framework that evaluates the performance of different optimization algorithms (artificial bee colony, ant colony optimization, differential evolution, particle swarm optimization, simulated annealing, satin bowered bird optimization, and genetic algorithm) to do feature selection on each of four benchmarked datasets (albrecht, cocomo, desharnais and maxwell) using artificial neural network, analogy based estimation, multiple regression, step wise regression and classification and regression tree as estimation methods. reduced number of dataset features besides the reduction of model complexity without loss of any estimation accuracy and data loss and also performance improvement, are the main advantages of the proposed framework.