Tackling Fingerprinting Indoor Localization Using the LASSO and the Conjugate Gradient Algorithms

Abstract

This paper presents the application and the comparison of the least absolute shrinkage and selection operator (LASSO) and the Conjugate Gradient (CG) algorithm for solving the fingerprinting indoor localization problem. LASSO's ability to generate sparsity via selection of variables results in a judicious and automatic removal of spurious measurements that often corrupt large fingerprint data sets. These spurious measurements usually have to be individually discarded before the CG algorithm, or other solver for the normal equation, is used. The paper also compares LASSO with a sparse version of the ordinary least squares solution obtained by simply discarding the variables with the smallest absolute value. The results are presented for two data sets recorded independently at the