As one kind of the multi-particulate dosage forms, the sustained-release pellets are becoming increasingly popular due to the desired medication safety and flexible control over drug release behavior [1]. Since pellets with different coating thickness or surface area tend to exhibit varying drug release rates, good quality coated sustained-release pellets should display an even coverage of coat, with smooth surface and no defects [2]. Besides, the roundness of pellets also affects the drug release. As postulated by Mosharraf and Nyström, elongated, more irregular particles could be surrounded by an on average thicker hydrodynamic boundary layer, through which the dissolved drug has to diffuse, extending the time of drug release [3]. The physical properties of the pellet cores significantly influence these properties of the coated pellets, assuming the stable preparation processes. It is conceivable that the pellet core with higher roundness exhibits more uniform coating thicknesses, which is conducive to achieving a more stable drug release. And the moisture contents of the pellet cores can affect coating effectiveness and dosing accuracy. Therefore, the knowledge and subsequent control of the physical properties of pellet cores is crucial in the development of sustained-release pellets.

The intrinsic properties of any powder are derived from five basic indices, including compressibility, flowability, particle size distribution, stackability and stability [4]. In the case of pellet core, in addition to the five indices, there is also the consideration of the pellet shape, that is, the roundness. The fingerprint technique has gained prominence in recent years as a valuable tool for assessing the comprehensive properties of samples [5]. For instance, chemical fingerprinting technology has been widely used for evaluating the quality of traditional Chinese medicines based on their chemical compositions [6]. In the pharmaceutical industry, the SeDeM expert system serves as a comprehensive tool for assessing the flow properties and compressibility of powders, aiding in the optimization of formulation and manufacturing processes [7], [8]. In order to comprehensively assess the physical characteristics of the pellet cores, the physical fingerprint, developed from the SeDeM expert system, has the potential to serve as an evaluation criterion. In general, physical fingerprints consist of more than 10 parameters, such as angle of repose, bulk density, bulk density, loss on drying, Carr’s index and particle size. However, the construction of physical fingerprints requires time-consuming and labor-intensive techniques and irrecoverably consumes a certain amount of sample, which making the rapid characterization of the physical properties of pellet core challenging. Therefore, a more suitable method is necessary to be developed to address this issue.

Due to the merits of rapid analysis, no complicated sample pretreatment and environmental friendliness, near-infrared (NIR) spectroscopy has extensive applications in the quantitative analysis of chemical compositions [9]. In the pharmaceutical industry, there has been growing interest in recent years regarding the potential application of NIR spectroscopy for the monitoring of physical parameters [10]. NIR chemical imaging and NIR spectroscopy can be used to detect the porosity distribution in the ribbons and the particle size of milled granules in dry granulation, as confirmed by the work of Khorasani [11]. Ortega-Zuniga et al. and Roman-Ospino et al. realized the real time monitoring of the powder density via NIR spectra [12], [13]. Additionally, there have been studies reporting the use of NIR spectroscopy technology for the characterization of flowability and compressibility properties [14], [15]. Chemometric data analysis is commonly required to achieve the correlation of physical parameters with NIR signals. Commonly employed algorithms for this purpose encompass direct standardization (DS), partial least squares regression (PLSR), and deep learning algorithms [16], [17], [18]. In the cases of linear issues, DS and PLSR are preferable choices, while the Generalized regression neural network (GRNN) algorithm excels in addressing nonlinear challenges [19], [20]. It is noteworthy that the correlation between the physical parameters of pellet core and NIR signals tends to be nonlinear. That is, the changes in the spectral absorption are more complexly related to the physical parameters and cannot be described by a simple proportional relationship. Therefore, it is necessary to investigate the potential of GRNN algorithm in physical fingerprint transformation. To improve the predictive capacity of the final model, ensemble strategy has evolved into a preferable option [21]. Bagging, also known as bootstrap aggregating, is one of the standard techniques for generating the ensemble-based algorithms [22]. The main idea in Bagging is to generate a series of independent observations with the same size, and distribution as that of the original data. Given the series of observations, generate an ensemble predictor which is better than the single predictor generated. This makes the model more stable and less susceptible to overfitting.

The physical information regarding pellet cores is crucial for the development of sustained-release pellets. Nonetheless, there is currently a deficiency in methods for the rapid characterization of pellet core physical properties. Thus, the objective of this study is to explore the application of NIR spectroscopy for transforming the physical fingerprint of pellet core and to compare the potential of DS, PLSR and GRNN algorithms based on Bagging.