Construction of software projects measurement result database and usage in new software projects' cost estimations

Cost estimation of computer software is getting more important. Software development becomes increasingly expensive. Software cost estimation is a very important problem for governments and organizations. There are a lot of projects which exceeded planned budged and time. Incorrect budget and time planning at the beginning is the main failure reason. Software becomes increasingly expensive to develop and is a major cost factor in any information system budget. Software development costs often get out of control due to lack of measurement and estimation methodologies. During last decade, some researches on cost estimation have been conducted. In the search of new methodologies to estimate the software development costs but the results were far from being satisfying. In this study, a new software engineering metric set was developed and the data was collected according to new metric set and a new software cost estimation model was developed by using neural network. In this study we have explored the reasons of the disappointing results of the existing software cost estimation with neural network studies and implemented different neural network models using augmented new metrics. The metric-set selection has a vital role in software cost estimation studies; its importance has been ignored especially in neural network based studies. The results obtained are compared with previous studies using traditional metrics. To be able to make comparisons, two types of data have been used. The first part of the data is taken from the Constructive Cost Model which is commonly used in previous studies and the second part is collected according to new metrics in a leading international company in Turkey. Another difficulty associated with the cost estimation studies is the fact that the data collection requires time and care. Futhermore, many companies do not share their data because of competition disadvantage possibilities. In recent literature, we have explored that MLP had been used in software cost estimation with neural network studies. In this study, software cost estimation by using neural network models presented here are based on Multi-Layer Perceptron (MLP) and Elman neural networks which has not been used for this aim. The results obtained are compared with previous studies using traditional metrics. The models presented here are based on Multi- Layer Perceptron and Elman Networks for both COCOMO?81 metric set and for the augmented metric set (YEEM : Yıldız Effort Estimation Metrics). To be able to compare the results accurately tests were run for datasets containing the same number of samples. To investigate further, we have also experimented with larger datasets formed according to augmented metrics. Addressing the issues of the dataset characteristics and the amount of samples in the datasets is one of the purposes of this research. YEEM has been used to construct the neural network topology. The MLP and Elman neural networks has been trainned and tested by using data for YEEM. To make a more thorough use of the samples collected, k-fold, cross validation method is also implemented. Since the amount of samples we have collected are still limited, 5-fold, 10 fold and 15-fold cross validation techniques have been also applied. It is concluded that, as long as an accurate and quantifiable set of metrics are defined and measured correctly, neural networks can be applied in software cost estimation studies with success.