Abstract:

Estimating effort early in the software development life cycle is essential for proper planning. It enables better allocation of resources, time, and budget, helping to avoid project delays and cost overruns. Inaccurate estimation often leads to project failures, which is a pervasive issue nowadays for software project managers. Machine learning approaches have generally shown significant success in addressing estimation challenges across various engineering domains. This study introduces a novel method, combining a Dense neural network (DNN) with a metaheuristic adaptive anti-predatory (AP) Algorithm known as AP-DNN. This method is effectively used to address the challenges of estimating software effort. The adaptive anti-predatory (AP) algorithm is utilized to optimize the parameters of the DNN, improving its capacity to explore the parameter space thoroughly and avoid getting trapped in local optima. The proposed anti-predatory dense neural network (AP-DNN) method was tested on several benchmark SEE datasets, and its performance was compared with various contemporary algorithms from existing literature. The comparative results indicate that AP-DNN outperforms other methods across most evaluation metrics and datasets.

Key words: estimation of software effort, anti-predatory NIA, memetic, optimization, levy flight, DNN