Nonoptimality Levels in Numerical Implementation of the Least Absolute Deviations Method
Alexander Matasov, Pavel A. Akimov
Anomalous measurement errors (outliers) are frequently occur in the processing of measurement data. In this case, the least absolute deviations (LAD) method is an effective estimation method. However, various numerical algorithms for the implementation of this method are iterative and the question of convergence rate is not always clear. The present paper is devoted to constructing the nonoptimality levels for the current iteration. These levels allow us to guarantee the approximation accuracy and in so doing to give a reliable stopping criterion for the iteration process.