A new method for detection of influential observations in linear regression models

Influential observations are observations that differ from other observations in the data set and cause a large change (effect) in the regression coefficients. The presence of such observations in the data set reduces the validity and sensitivity of statistical analyzes. There are many methods used to determine influential observations in the literature. However, most of these methods require distributional assumptions and are highly affected by masking and swamping effects. Most of these methods are inadequate, especially if there is more than one influential observation in the data set. The aim of this thesis is to develop a new diagnostic method to detect influential observation sets using the meta-heuristic Binary Particle Swarm Optimization algorithm. This proposed approach does not require any distributional assumptions and is also not affected by the masking and swamping effects such as the classical methods. The performance of the proposed method has been analyzed via various simulations and real data set applications.