Optimization of neural network for software effort estimation

The effort required for the development of a software system is predicted through the cost of software estimation. Completion of project within time and budget limits is required for accurate cost estimation. Effort and cost estimation can be done through various modes. A new hybrid algorithm which is a combination of concepts of Artificial Bee Colony (ABC) and Local search procedures is used here. MultiLayer Perceptron Neural Network (MLPNN) with hybrid algorithm is presented in this paper to improve software effort estimation. The proposed protocol is evaluated through Constructive Cost Model (COCOMO) dataset. Mean Magnitude Relative Error (MMRE) and Median Magnitude Relative Error (MdMRE) are the criteria used for evaluation. COCOMO dataset attributes are transformed using Principal Component Analysis (PCA) and ranked through a Correlation based Feature Selection (CFS). The proposed hybrid MLPNN classifies the ranked attributes.