Optimization of Cost Estimation Models using Soft Computing Techniques

Abstract

Estimations produced by algorithmic cost models needs improvement in terms of accuracy. In the proposed model, inference system based on fuzzy logic is designed for calculating the value of impact of each cost driver. The traditional regression based Intermediate COCOMO model considers the impact of fifteen drivers based on multiple factors like product related, project related, person related and platform related. Using fuzzy inference system, the input related to each cost driver is represented in terms of fuzzy sets. The modification suggested have been verified by validating the proposed model using two different datasets COCOMO NASA and COCOMO NASA2.The research finding includes a comparative analysis of the Mean Magnitude of Relative Error MMRE and Prediction accuracy evaluation criteria for the proposed model. Incorporating fuzzy logic improves the overall estimation accuracy. newlineIn the proposed model, hybridization of fuzzy modelling and genetic algorithm is implemented in the form of genetic fuzzy system. In the proposed hybrid model, tuning of fuzzy sets is done using genetic algorithms. As a result estimation model has reduced error function and hence improved accuracy level. newlineAnother component of evolutionary computing which can also be implemented to solve optimization problem is swarm intelligence. In the proposed methodology, PSO is used to optimize the process of generating parameters for fuzzy sets. Each particle represents a set of membership function for each cost driver of Intermediate COCOMO. The results obtained in terms of MMRE shows an improvement over fuzzy model justifying the proposed optimization task. newlineThe performance of all the three proposed models; Fuzzy Model For Intermediate COCOMO, GA optimized Fuzzy Model for Intermediate COCOMO and PSO optimized Fuzzy Model for Intermediate COCOMO have been tested using two NASA datasets available in promise repository. newline newline newline