The present study explores the effect of soil-structure interaction (SSI) on the foundation motion as recorded at accelerometric stations located at the basement of buildings through a large dataset employed from sites where both foundation and free-field ground motion recordings are available. The significance of such instrumentation is highlighted, in terms of assessing the high frequency filtering that occurs at the foundation level and its potential implication on ground motion prediction models (GMPMs) that are produced with these datasets. Based on the recorded data, analytical expressions relating foundation, and free-field motion intensity measures, which have been produced in a previous work, are verified. A sub-structure analysis procedure is extended to include the site and building characteristics of the recorded data. Nonlinear regression analyses are performed, utilizing numerous analysis results, to derive improved, robust analytical expressions, as well as, to include building characteristics. Residual analysis is performed to assess possible bias due to other variables, as well. The produced analytical expressions are validated against the recorded data and compared to the existing expressions in terms of prediction errors. The produced analytical expressions can be utilized for correcting motions recorded at the basement level of buildings to obtain estimates of free-field ground motions. The latter being "building-free" is most appropriate for the development of new generation GMPMs that are not biased by the influence of kinematic and inertial interaction. The impact of correcting ground motion data through the proposed analytical expressions is discussed through their implementation on recently published strong motion dataset.