n-Steps ahead software reliability prediction using the Kalman filter

This paper presents KSL, a new software reliability growth model (SRGM) based on the Kalman filter with a sub filter and the Laplace trend test. We applied the model to the Linux operating system kernel as a case study to predict the absolute and relative (per lines of code) number of faults n-steps ahead. The Laplace trend test is applied to detect when the series no longer follows a homogeneous Poisson process, improving the confidence level. An example is provided with a prediction of 13 months ahead on the number of faults with 8% error. The results (i.e. predictive capability) indicated that the proposed approach outperforms the S-shaped prediction model, Weibull, and Exponentiated Weibull distributions, as well as typical and OS-ELM Neural networks when the series has a short number of observations.