Error Prediction Methods for Embedded Software Development Using Hybrid Models of Self-Organizing Maps and Multiple Regression Analyses

Abstract

In this study, we establish error prediction models at various stages of embedded software development using hybrid methods of self-organizing maps (SOMs) and multiple regression analyses (MRAs). SOMs are a type of artificial neural networks that relies on unsupervised learning. A SOM produces a low-dimensional, discretized representation of the input space of training samples; these representations are called maps. SOMs are useful for visualizing low-dimensional views of high-dimensional data as a multidimensional scaling technique. The advantages of SOMs for statistical applications are as follows: (1) enabling reasonable inferences to be made from incomplete information via association and recollection, (2) visualizing data, (3) summarizing large-scale data, and (4) creating nonlinear models. We focus on the first advantage to create error prediction models at various stages of embedded software development. In some cases, a model using only SOMs yields lower error prediction accuracy than a model using only MRAs. However, the opposite is true. Therefore, in order to improve prediction accuracy, we combine both models. To verify our approach, we perform an evaluation experiment that compares hybrid models to MRA models using Welch’s t test. The results of the comparison indicate that the hybrid models are more accurate than the MRA models for the mean of relative errors, because the mean errors of the hybrid models are statistically significantly lower.