NIR spectroscopy has been extensively employed for the in-line monitoring of pharmaceutical processes as one of the key PAT implementation tools. Nevertheless, pharmaceutical processes such as fluid-bed coating have not fully made the most of the NIR in-line monitoring primarily due to a difficulty in handling random in-line spectra. In this study, novel approaches to develop a reasonable dynamic calibration model were proposed; averaging and clustering. Pharmaceutical test tablets were coated with HPMC-based materials using a fluid-bed processor. During the 160 min coating process under tangential spraying mode, 10 tablets were sampled out at every 10 min mark for actual coating thickness measurements. NIR spectra at and near each 10 min mark were treated and processed by the averaging and clustering operations. Averaging of 21 spectra resulted in a reasonably good dynamic calibration model whose determination coefficient was estimated as high as 0.9916. The PCA-based clustering turned out to be substantially helpful especially when a large number of NIR spectra were averaged. A prediction experiment verified that our dynamic calibration model can control the coating thickness in-line as good as 3% deviated from the actual thickness, which can offer a reasonable end-point for the fluid-bed coating process.