Incremental regularized Data Density-Based Clustering neural networks to aid in the construction of effort forecasting systems in software development

摘要

The challenge of reducing complexity, failures and time in software development are tasks found in the vast majority of information technology companies. The professionals seek to structure the development of applications through agile methodologies, but in reality, there are several difficulties in planning the time to make applications development tasks. To help in this situation, this paper proposes the use of fuzzy neural network composed by fuzzy rules to assist in the construction of a specialist system based on interpretable rules, facilitating the prediction of software development hours according to the complexity of the elements present in the project. To support in the data fuzzification process, an incremental density technique is proposed for the first layer of the model. To improve the sensitivity of the neuron present in the neural network a Leaky-ReLU type activation function is used to obtain the results. The set of rules to be created, through tests in a real database based on the technique of use case point, can help in the development of future expert systems, to be used by these professionals. The results of the tests were efficient to generate predictability about the efforts to build the software.