Improving Accuracy of an Artificial Neural

Improving Accuracy of an Artificial Neural
Network Model to Predict Effort and Errors in
Embedded Software Development Projects
Kazunori Iwata, Toyoshiro Nakashima, Yoshiyuki Anan, and Naohiro Ishii
Abstract. In this paper we propose a method for reducing the margin of error in effort
and error prediction models for embedded software development projects using
artificial neural networks(ANNs). In addition, we perform an evaluation experiment
that uses Welch’s t-test to compare the accuracy of the proposed ANN method with
that of our original ANN model. The results show that the proposed ANN model
is more accurate than the original one in predicting the number of errors for new
projects, since the means of the errors in the proposed ANN are statistically significantly
lower.
1 Introduction
Due to the expansion in our information-based society, an increasing number of information
products are being used. In addition the functionality thereof is becoming
Kazunori Iwata
Dept. of Business Administration, Aichi University
370 Shimizu, Kurozasa-cho, Miyosh, Aichi, 470-0296, Japan
e-mail: kazunori@vega.aichi-u.ac.jp
Toyoshiro Nakashima
Dept. of Culture-Information Studies, Sugiyama Jogakuen University
17-3, Moto-machi, Hoshigaoka, Chikusa-ku, Nagoya, Aichi, 464-8662, Japan
e-mail: nakasima@sugiyama-u.ac.jp
Yoshiyuki Anan
Omron Software Co., Ltd.
Shiokoji Horikawa, Shimogyo-ku, Kyoto, 600-8234, Japan
e-mail: y-anan@mx.omronsoft.co.jp
Naohiro Ishii
Dept. of Information Science, Aichi Institute of Technology
1247 Yachigusa, Yakusa-cho, Toyota, Aichi, 470-0392, Japan
e-mail: ishii@aitech.ac.jp
Roger Lee (Ed.): SNPD 2010, SCI 295, pp. 11–21, 2010.
springerlink.com c Springer-Verlag Berlin Heidelberg 2010
12 K. Iwata et al.
ever more complex[3, 8]. Guaranteeing software quality is particularly important,
because it relates to reliability. It is, therefore, increasingly important for embedded
software-development corporations to know how to develop software efficiently,
whilst guaranteeing delivery time and quality, and keeping development costs low
[2, 6, 7, 9, 10, 12, 13, 14]. Hence, companies and divisions involved in the development
of such software are focusing on a variety types of improvements, particularly
process improvement. Predicting manpower requirements of new projects and guaranteeing
quality of software are especially important, because the prediction relates
directly to cost, while the quality reflects on the reliability of the corporation. In
the field of embedded software, development techniques, management techniques,
tools, testing techniques, reuse techniques, real-time operating systems, and so on,
have already been studied. However, there is little research on the relationship between
the scale of the development, the amount of effort and the number of errors,
based on data accumulated from past projects. Previously, we investigated the prediction
of total effort and errors using an artificial neural network (ANN) [4, 5].
In earlier papers, we showed that ANN models are superior to regression analysis
models for predicting effort and errors in new projects. In some projects, however,
the use of an ANN results in a large margin for error. In this paper, we propose a
method for reducing this margin of error and compare the accuracy of the proposed
method with that of our original ANN model.