WCET estimation using support vector regression based on Legendre orthogonal kernel functions

In computer science, particularly in real-time systems, due to the existence of deadlines for executable tasks, a concept called worst-case execution time (WCET) has played an important role. Determining WCET is essential for effectively scheduling tasks, thus estimating this parameter is highly important. Due to the complexities of software and hardware, there are many factors affecting the execution of a task, so there are many challenges to estimating WCET. Due to these challenges, nowadays machine learning methods such as support vector regression (SVR) are employed to estimate WCET. In addition, orthogonal functions are utilized as an effective way to use as a kernel function due to the existence of minimal data redundancy in the feature space. In this research, the Legendre orthogonal functions have been used as the SVR kernel in order to estimate WCET. To examine the experimental results of this research, Mälardalen University benchmarks were used, and the results of the proposed method improving the accuracy and tightness of WCET in 69% of the benchmarks were obtained. Furthermore, in 69% of the benchmarks, a safeness percentage of over 60% was achieved.