A Real Time Extreme Learning Machine for Software Development Effort Estimation

Software development effort estimation always remains a challenging task for project managers in a software industry. New techniques are applied to estimate effort. Evaluation of accuracy is a major activity as many methods are proposed in the literature. Here, we have developed a new algorithm called Real Time Extreme Learning Machine (RT-ELM) based on online sequential learning algorithm. The online sequential learning algorithm is modified so that the extreme learning machine learns continuously as new projects are developed in a software development organization. Performance of the real time extreme learning machine is compared with training and testing methodology. Studies were also conducted using radial basis function and additive hidden node. The accuracy of the Real time Extreme Learning machine with continuous learning is better than the conventional training and testing method. The results also indicate that the performance of radial basis function and additive hidden nodes is data dependent. The results are validated using data from academic setting and industry.