Since the fourth industrial revolution, artificial intelligence has been used in various fields. Accordingly, there is an increasing demand for calculating software costs in a systematic and standard manner through the application of artificial intell...
Since the fourth industrial revolution, artificial intelligence has been used in various fields. Accordingly, there is an increasing demand for calculating software costs in a systematic and standard manner through the application of artificial intelligence. Currently, software development is increasing in complexity due to universalization of service delivery methods using various devices such as mobile devices. In addition, the scale and complexity of software has grown in response to increasing demand for an environment in which work can be done anytime, anywhere. This has made it difficult for experts to guarantee the reliability of cost calculations, as differences arise in the results of cost estimates. In general, calculating software cost estimation, including software development and operation, is common in FP (Function Point) and M/M (Man/Month) methods for calculating input personnel based on workload. Among them, this paper discusses the FP method, which is relatively more reliable. Currently, it is recommended that public agencies' software cost estimates be calculated using the KOSA (Korea Software Industry Association) "Software Cost Estimation Guide" model. However, the subjective criteria used by cost estimators can lead to over- or underestimation. There are several cases in which there have been discrepancies between cost-assessment and the actual costs of a project due to the vague criteria used by cost-estimation experts.
First, this paper proposes cost estimation modeling based on artificial intelligence. Second, the K-Means clustering algorithm is applied to the "Software Cost Estimation Guide" model, which is the application standard for public institutions. Third, the attributes needed for the model are defined based on the analyzed modeling results and orthodoxy, which is the FP calculation method. Finally, the proposed modeling is evaluated by applying a linear regression algorithm. The evaluation result shows 80% accuracy, which indicates that the prediction model works well. If AI learning is carried out by continuously acquiring a large amount of data in the future, it is expected that it can be effectively and efficiently applied to calculating cost estimations at industrial sites. The proposed model for predicting software costs is expected to enable the calculation of more standardized and reliable cost estimations than the "Software Cost Estimating Guide" model.