روش استخراج قانون رابطه برای برآورد تاثیر سیاست های مدیریت پروژه بر کیفیت نرم افزار، زمان توسعه و تلاش
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|15704||2008||8 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Expert Systems with Applications, , Volume 34, Issue 1, January 2008, Pages 522-529
Accurate and early estimations are essential for effective decision making in software project management. Nowadays, classical estimation models are being replaced by data mining models due to their application simplicity and the rapid production of profitable results. In this work, a method for mining association rules that relate project attributes is proposed. It deals with the problem of discretizing continuous data in order to generate a manageable number of high confident association rules. The method was validated by applying it to data from a Software Project Simulator. The association model obtained allows us to estimate the influence of certain management policy factors on various software project attributes simultaneously.
Software quality, project duration and development effort are important factors to be kept under control in the software development process. They are interrelated and influenced by many other factors which complicate the monitoring tasks. When managers have to take decisions about a project, they must consider a great number of variables and the complex relations between them. The simulation of software development projects by using dynamic models has contributed to a better knowledge of the influence of these variables and their relations. The Software Project Simulators (SPS), based on dynamic models, enable us to simulate the project’s behavior and to evaluate the impact of different management policies and other factors. Nevertheless, they have important drawbacks: first, the number of input parameters needed for the simulation, and second, the number of possible combinations of factors influencing the development process that make it difficult to choose the best combination for the desired objectives. An important improvement in SPS is the treatment of the data generated by the simulator by using machine learning and evolutionary algorithms in order to facilitate their use ( Aguilar-Ruiz et al., 2001 and Ramos et al., 2001). The combination of factors for achieving specific objectives can be learned through the application of these supervised techniques. Such information allows managers to establish the correct management policy taking as reference the model generated by these algorithms. When machine learning techniques are used, only one output variable is the target of the prediction. In classification problems this variable, named the class attribute, must be discrete. Classification algorithms use classified historical data for inducing a relation model between the class attribute and the other attributes (descriptive attributes). Later, the model can make predictions about new, unclassified data. The aim of this work is to present a method for estimating three variables simultaneously. We propose an association rule mining algorithm for building a model that relates management policy attributes to the output attributes quality, time and effort. All the available attributes are continuous, they must thus be split into intervals of values in order to generate the rules. The applicability and interest of the discovered associations depend mostly on how the data is discretized. The success of our method is mainly due to the supervised, multivariable procedure used for discretization. The result is an association model comprised of a manageable number of high confidence rules representing relevant patterns between project attributes. Those patterns provide managers with important information for decision making. The rest of the paper is organized as follows: next section contains the fundamentals and main works concerning SPS, association rules and data discretization. Section 3 describes the experimental data provided by a dynamic simulation environment. The following section deals with the stage of data preprocessing. The proposed method for association rule mining and its results are presented in Section 5. The evaluation of the associative model obtained is given in Section 6 and, finally, we draw some conclusions.
نتیجه گیری انگلیسی
Some management policy factors have a great influence on the success of a software project; however it is very difficult to know their impact on other project attributes, due to the complex relations existing between them. In this paper we have presented a data mining study of this influence by using data from an SPS based on a dynamic model. We have proposed an association rule mining algorithm for building a model that relates management policy attributes with the output attributes quality, time and effort. The success of the algorithm is mainly due to the supervised multivariate procedure used for discretizing the continuous attributes in order to generate the rules. The result is an association model comprised of a manageable number of high confidence rules representing relevant patterns between project attributes. This allows us to estimate the influence of the combination of some variables related to management policies on software quality, project duration and the development effort simultaneously. Classical machine learning methods can only predict one variable at a time. The study has demonstrated that the delay in the departure of the new technicians and the maximum allowed percentage of delivery time with regard to the initially estimated time have no appreciable influence on the studied attribute projects. However, factors relating to the incorporation and adaptation of the new technicians have an important impact, as the associative model obtained shows. In addition, the proposed method avoids three of the main drawbacks of rule mining algorithms: production of a high number of rules, discovery of uninteresting patterns and low performance.