This paper presents an enhancement of a software project simulator to perform risk prediction with cost estimation capability. So far, we have developed a software project simulator to simulate software development projects. In this simulator, a development process was described using Petri net model, and it was applied to some actual project data in a certain company successfully. On the other hand, we have also presented a risk predicting system to find "risky" projects by statistical analysis on risk questionnaire for project managers. In this approach, only the probability to be risky was calculated for a project. Thus, the managers in the company wanted to know a concrete proof why a software project becomes risky. In this paper, to present the proof that a software project becomes risky, we try to enhance the previous project simulator so that the simulator can deal with risk factors. To consider the risk factors, we modify the previous simulator so that both the fluctuation of skill level and the deadline pressure can be represented by the parameters in the simulator. By using a case study, we confirm that the enhanced simulator can estimate the development cost under some typical risks. As a result, we can expect that the simulator shows how much the development cost of a risky project exceeds an estimate.

In this paper, we consider a simple development process consisting of design and debug phases, which is derived from actual concurrent development process for embedded software at a certain company. Then we propose two-phase project control that examines the initial development plan at the end of design phase, updates it to the current status of the development process and executes the debug phase under the new plan. In order to show the usefulness, we define three imaginary projects based on actually executed projects in a certain company: the project that executes debug phase under initial plan, the project that applies the proposed approach, and the project that follows a uniform plan. Moreover, to execute these projects, we use the project simulator, which has already been developed based on GSPN model. Judging from the number of residual faults in all products, we found that project B is the best among them.

Recently, software development projects have been required to produce highly reliable systems within a short period and with low cost. In such situation, software quality prediction helps to confirm that the software product satisfies required quality expectations. In this paper, by using a Bayesian Belief Network (BBN), we try to construct a prediction model based on relationships elicited from the embedded software development process. On the one hand, according to a characteristic of embedded software development, we especially propose to classify test and debug activities into two distinct activities on software and hardware. Then we call the proposed model "the BBN for an embedded software development process". On the other hand, we define "the BBN for a general software development process" to be a model which does not consider this classification of activity, but rather, merges them into a single activity. Finally, we conducted experimental evaluations by applying these two BBNs to actual project data. As the results of our experiments show, the BBN for the embedded software development process is superior to the BBN for the general development process and is applicable effectively for effective practical use.