1. Please check and comment entries here.
Table of Contents

    Topic review

    Machine Learning for Process Monitoring

    View times: 7
    Contributors: ,
    Submitted by:


    In the last few decades, hot-melt extrusion (HME) has emerged as a rapidly growing technology in the pharmaceutical industry, due to its various advantages over other fabrication routes for drug delivery systems. After the introduction of the ‘quality by design’ (QbD) approach by the Food and Drug Administration (FDA), many research studies have focused on implementing process analytical technology (PAT), including near-infrared (NIR), Raman, and UV–Vis, coupled with various machine learning algorithms, to monitor and control the HME process in real time. This review gives a comprehensive overview of the application of machine learning algorithms for HME processes, with a focus on pharmaceutical HME applications. The main current challenges in the application of machine learning algorithms for pharmaceutical processes are discussed, with potential future directions for the industry.

    1. Introduction

    Hot-melt extrusion (HME) is a rapidly growing technology in the pharmaceutical industry, for the preparation of various dosage forms, including granules, pellets, tablets, and implants. The HME process offers many advantages relative to other pharmaceutical processes, one of the major benefits being that HME can enhance the bioavailability and solubility of poorly soluble drugs. Further, as a solvent-free process, it is free of harsh environmental toxicants and no additional step for solvent recovery is required, unlike solvent evaporation and spray drying. HME is also relatively easy to scale-up, and it is a continuous process [1].
    As with all pharmaceutical products, polymer–drug extrudates that are produced using HME must undergo rigorous quality analysis and typically undergo thermal, rheological, mechanical, and chemical characterisation. For thermal analysis, DSC and TGA are widely used to measure the percentage of crystallinity, the glass transition temperature (Tg), and the change in weight. The results of these methods have been used to predict the miscibility, solid state, and stability of the polymer–drug matrix [2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19]. Rheological analysis is used to provide information about the behaviour of the polymer–drug system under the high temperature and stresses experienced in the process itself [20][21][22][23]. HPLC is used to monitor the drug/additive content in the extrudate [4][5][14][18][19]. FT-IR, Raman, and NIR have been employed post-production to study the stability and for analysing the drug content in the polymer–drug matrix [2][4][10][18][19][24]. Physicomechanical testing [25][26] and dissolution testing [27][28] are also performed for product quality assurance. The main disadvantage of these off-line, lab-based methods is that there is a long lag between processing and feedback on product quality, which makes process control very challenging. HME is a continuous process, but the long testing time for product quality assurance defeats this advantage of the process.
    In 2004, the Food and Drug Administration (FDA) introduced the concept of using process analytical technology (PAT) [29]. The main aim is to improve the understanding of the mechanism of the manufacturing process, enhance process monitoring, and to reduce the processing time. In the literature, spectroscopic techniques, including Raman [2], NIR [30], and UV–Vis spectroscopy [31], have been widely implemented as PAT tools for in/on-line monitoring of the HME process, and an in-line slit die rheometer has also been implemented as a PAT tool in some studies [32][33]. Machine learning (ML) algorithms are generally used to infer the required information from in/on-line collected spectra.
    A PAT tool, coupled with a machine learning algorithm, has been established as an effective way to monitor the HME process in real-time. Figure 1 gives a schematic representation of the work flow for in/on-line monitoring of the HME process using PAT tools coupled with machine learning. Since 2004, many research studies have been reported, in which different machine learning algorithms have been applied to in-process data to analyse product and process parameters in real-time. The applications include the monitoring of product critical quality attributes (CQAs), including the following: the solid state of the polymer/drug [2][3]; API/additive concentration [34][35]; degradation of the polymer [36][37]; the particle size of additive/s [31]; and mechanical properties [33]. Other works have examined the monitoring of critical process properties (CPPs), including melt temperature [38], pressure [39], and viscosity [40], and for process fault detection [41].
    Figure 1. Schematic representation of in/on-line monitoring of HME process with machine learning.
    In recent years, many review papers have been published focusing on different aspects of the HME process [1][15][42][43][44][45][46][47][48][49][50][51][52][53][54][55][56][57][58]. In this review, we focus specifically on the application of machine learning (ML) algorithms in the monitoring and control of the HME process. Greater process sensorization, coupled with algorithms for deriving intelligence from data, are key concepts of the Pharma 4.0 initiative for the digital transformation of the pharmaceutical industry, hence this review aims to establish the current state of the art of the HME process in this respect. We present and discuss the various data analytics and machine learning methods reported for the monitoring and control of HME, including the following: methods for the pre-processing of data, model training/calibration, the ability of the developed model to detect the effect of varying processing conditions, and the performance of models on unseen data. We summarise the contribution that machine learning has made to date in the monitoring and control of the HME process and discuss the main challenges and future potential of the field.
    The remainder of the paper is organised as follows: First, a brief introduction is given to machine learning and the main data pre-processing techniques that are relevant to the HME process. The main body of the paper then reviews the applications of (i) PCA, (ii) PLS, and (iii) non-linear machine learning algorithms to the process, followed by the discussion and conclusions.

    2. Machine Learning

    Machine learning (ML) is generally defined as the ability of a computer to learn without being explicitly programmed. Machine learning algorithms train themselves to identify patterns in the data or make predictions based on past data, as opposed to modelling algorithms that are based on the prior physical/chemical knowledge of a system. A machine learning system can be predictive; descriptive (meaning that the system uses the data to explain what happened); or prescriptive (meaning that the system will use the data to make suggestions about what action to take). ML algorithms can be divided into the following three classes: supervised learning, unsupervised learning, and reinforcement learning [59].

    2.1. Supervised Machine Learning

    In supervised learning, algorithms are provided with known/labelled input–output data [59][60]. In other words, supervised machine learning algorithms try to predict the results for an unknown output based on the patterns present in the labelled data set, i.e., the algorithm tries to approximate the mapping function from input to output variables. Regression and classification are categorised under supervised machine learning.
    Classification algorithms classify training data into separate categorical classes/groups. All the samples of data in the training set are labelled. The purpose of using classification is to identify the class of future unknown observations. There are the following three types of classification: binary classification with two possible outcomes; multi-class classification with more than two classes; and multi-label classification, whereby each input in the training data is mapped to more than one class [61]. The classification algorithm’s performance is assessed based on how well an algorithm classifies unseen observations into the correct classes. A confusion matrix is created for performance assessment, where the rows represent the true classes and the columns represent the predicted classes. Naïve Bayes, k-nearest neighbours (k-NN), decision tree, support vector machine (SVM), and random forest (RF) are commonly used classification algorithms [62][63].
    In regression, the class of the output variable is continuous numeric. Linear regression methods include methods such as partial least squares regression (PLS), least absolute selection shrinkage operator (LASSO), and ridge regression; while random forest (RF) regression and support vector regression (SVR) are commonly used non-linear regression algorithms. In the literature, the performance of a regression algorithm is generally assessed based on its root mean square error (RMSE), which is based on the difference between the actual and predicted values, and on the coefficient of correlation (R2) values.

    2.2. Unsupervised Machine Learning

    Unlike supervised learning, the inputs are not labelled in unsupervised learning and the algorithm is concerned with detecting regularities/patterns in the unlabelled training data [59]. Clustering (e.g., k-means and hierarchical clustering) is a well-known class of unsupervised machine learning [60][64]. In clustering, the aim is to find similar subgroups within the data set; all the objects are divided into a certain number of clusters, and inputs with a similar pattern are gathered in the same cluster. Principal component analysis (PCA) is another very common unsupervised machine learning method. It is usually used for dimensionality reduction in data sets with a degree of collinearity between the input variables [65]. In PCA, the input variables are transformed into a new set of input features, which are linear combinations of the original variables. These new features or ‘principal components’ (PCs) successively explain the variance in the input data, such that most of the variation in the data can be captured by a small number of PCs and redundant input features, representing noise in the data set, can be ignored.

    2.3. Reinforcement Learning

    In reinforcement learning (RL), the learning process is different from supervised and unsupervised learning. Reinforcement learning is an agent-based learning process, whereby a ‘reward’ is associated with each learning action by the agent. An RL process proceeds with trial and error, and an agent learns through its interaction with the environment. To achieve the given task, i.e., to maximise the reward signal, it takes different actions, and experiences many failures and successes [66][67].

    3. Pre-Processing Techniques for In-Process Spectral Data

    Raw spectral data, collected using spectroscopic methods, typically undergoes pre-processing before applying a chemometric model [68]. During the process, spectral data can be affected by nuisance factors, including physical interruptions and faulty apparatus. These factors can reduce the signal-to-noise ratio and resolution [69]. Other undesirable features of raw spectra are baseline shifts and a complex background. Baseline shifts are caused by the scattering of the light, resulting from the interaction of spectra with the sample particles [68]. Undesired scatter effects can dominate the desired information (e.g., chemical information) in the spectra [70]. These undesired spectral variations can increase the complexity and reduce the accuracy of the model [69]. The main goal of pre-processing techniques is to remove the unwanted features from the spectra.
    For in-process spectral data, the following two groups of pre-processing techniques dominate the literature: scatter correction and spectral derivatives. Multiplicative scatter correction (MSC), extended MSC (EMSC), extended inverse MSC, de-trending, normalisation, and standard normal variate (SNV) belong to the scatter-correction group; these methods are used to correct baseline shifts and trends in the baseline. The spectral derivative group includes Norris-Williams (NW) and Savitzky-Golay (SG) and are used for smoothing and for reducing the noise effects [68][71]. The most common pre-processing techniques used in the literature are MSC, SNV, derivatives, and SG.
    MSC is used to remove undesired scatter effects. MSC defines a reference spectrum, which is commonly the average spectrum of the calibration set [72]. MSC is a two-step process involving the estimation of shifting and scaling correction coefficients [68]. After MSC, all the spectra have the same offset and amplitude [68][70]. SNV is also used to eliminate baseline shifts. SNV and MSC are quite similar to each other, but in SNV, a spectrum is mean-centered and then scaled by its standard deviation [68]. Smoothing is also a pre-processing method used to increase the signal-to-noise ratio. The moving average, where each spectral point is substituted by the average of m neighbouring points, is the simplest smoothing method (m is defined as the width of the smoothing window) [71]. Savitzky-Golay (SG) is a popular smoothing method that performs local least squares regression on the spectral data [72][73]. Differentiation is usually applied after applying smoothing methods. Derivatives are used to increase the spectral resolution and to eliminate the background effects. The first derivative eliminates constant baseline shifts, while the second derivative eliminates linear shifts in the spectrum, along with eliminating constant baseline shifts [72][73][74]. For more in-depth reading on these methods to understand the differences and similarities, the reader is directed to the review article [68] and these papers [69][71][75].

    4. Application of PCA for In-ProcessMonitoring of Critical Quality Attributes (CQAs)

    PCA is a technique for dimensionality reduction, which falls under unsupervised machine learning [65,77,78]. Details on how PCA works with PAT tools for monitoring pharmaceutical processes (other than HME) can be found in [79–81], and the detail of the algorithm is not repeated here. PCA has mostly been utilised in the HME literature to monitor the effect of varying processing conditions on the solid state of the drug. The drug solid state significantly influences the dissolution rate and bioavailability of the drug, with an amorphous form of the drug exhibiting a higher dissolution rate than the crystalline form.

    5. Application of PLS for In-Process Monitoring of Critical Quality Attributes (CQAs)

    PLS regression is a multivariate linear regression method that is suitable for highly collinear data. Analogous to PCA, it involves a linear transformation of the data set, allowing for dimensionality reduction to a reduced number of ‘latent variables’ (LV), which are linear combinations of the original variables. General details on the workings of the PLS algorithm applied to PAT data for pharmaceutical process monitoring can be found in [79–81,84]. In pharmaceutical processes, PLS is primarily used to predict the concentration of the drug, although it has also been used to predict polymer blend contents, degradation of the polymer, the particle size of fillers in the polymer matrix, and mechanical properties of the polymer extrudate in non-pharma HME processes.

    6. Conclusions

    The application of machine learning in pharmaceutical processing is a rapidly developing field, with many potential benefits for process optimisation and control. A well-designed machine learning model can speed up the development process, aid optimisation of the process, reduce the process cost, enhance product consistency, reduce process faults, and enable rapid validation of product quality. However, the use of machine learning algorithms for pharmaceutical HME is relatively new and is as yet underdeveloped. Most of the works reported in the literature have been conducted to predict/monitor the solid state of the polymer–drug extrudate and the API concentration. Few works have been published to date on predicting the final properties of the product such as, degradation of the polymer–drug matrix, mechanical properties and rate of loss of mechanical properties, drug release profile, etc., from in-process data. Recent works examining the application of more-complex ML models, both in HME and more widely in pharmaceutical processing, indicate that with careful design of the sensing system, the experimental procedures, and the modelling algorithms that prediction of such properties from in-process data may be possible in the future. Further, the application of machine learning for automating process control, for example, by using reinforcement learning, has not yet been explored in the literature. Future work should be in the direction of examining the suitability of different machine learning methods, their robustness, and limitations to predict and control the final properties of the polymer–drug matrix. It is stressed that if such models are to meet the industrial requirements for product validation that appropriately rigorous model validation procedures should be applied 



    This entry is adapted from 10.3390/pharmaceutics13091432


    1. Tiwari, R.V.; Patil, H.; Repka, M.A. Contribution of hot-melt extrusion technology to advance drug delivery in the 21st century. Expert Opin. Drug Deliv. 2016, 13, 451–464.
    2. Saerens, L.; Ghanam, D.; Raemdonck, C.; Francois, K.; Manz, J.; Krüger, R.; Krüger, S.; Vervaet, C.; Remon, J.P.; De Beer, T. In-line solid state prediction during pharmaceutical hot-melt extrusion in a 12 mm twin screw extruder using Raman spectroscopy. Eur. J. Pharm. Biopharm. 2014, 87, 606–615.
    3. Van Renterghem, J.; Kumar, A.; Vervaet, C.; Remon, J.P.; Nopens, I.; Vander Heyden, Y.; De Beer, T. Elucidation and visualization of solid-state transformation and mixing in a pharmaceutical mini hot melt extrusion process using in-line Raman spectroscopy. Int. J. Pharm. 2017, 517, 119–127.
    4. Psimadas, D.; Georgoulias, P.; Valotassiou, V.; Loudos, G. Molecular Nanomedicine towards Cancer. J. Pharm. Sci. 2012, 101, 2271–2280.
    5. Thiry, J.; Lebrun, P.; Vinassa, C.; Adam, M.; Netchacovitch, L.; Ziemons, E.; Hubert, P.; Krier, F.; Evrard, B. Continuous production of itraconazole-based solid dispersions by hot melt extrusion: Preformulation, optimization and design space determination. Int. J. Pharm. 2016, 515, 114–124.
    6. Huang, S.; O’Donnell, K.P.; Delpon de Vaux, S.M.; O’Brien, J.; Stutzman, J.; Williams, R.O. Processing thermally labile drugs by hot-melt extrusion: The lesson with gliclazide. Eur. J. Pharm. Biopharm. 2017, 119, 56–67.
    7. Li, Y.; Pang, H.; Guo, Z.; Lin, L.; Dong, Y.; Li, G.; Lu, M.; Wu, C. Interactions between drugs and polymers influencing hot melt extrusion. J. Pharm. Pharmacol. 2014, 66, 148–166.
    8. Crowley, M.M.; Zhang, F.; Koleng, J.J.; McGinity, J.W. Stability of polyethylene oxide in matrix tablets prepared by hot-melt extrusion. Biomaterials 2002, 23, 4241–4248.
    9. Backes, E.H.; Pires, L.D.N.; Costa, L.C.; Passador, F.R.; Pessan, L.A. Analysis of the Degradation During Melt Processing of PLA/Biosilicate® Composites. J. Compos. Sci. 2019, 3, 52.
    10. Guo, Z.; Lu, M.; Li, Y.; Pang, H.; Lin, L.; Liu, X.; Wu, C. The utilization of drug-polymer interactions for improving the chemical stability of hot-melt extruded solid dispersions. J. Pharm. Pharmacol. 2014, 66, 285–296.
    11. Huang, S.; O’Donnell, K.P.; Keen, J.M.; Rickard, M.A.; McGinity, J.W.; Williams, R.O. A New Extrudable Form of Hypromellose: AFFINISOLTM HPMC HME. AAPS PharmSciTech 2016, 17, 106–119.
    12. Haser, A.; Huang, S.; Listro, T.; White, D.; Zhang, F. An Approach for Chemical Stability during Melt Extrusion of a Drug Substance with a High Melting Point; Elsevier: Amsterdam, The Netherlands, 2017; Volume 524, ISBN 5124710942.
    13. Hengsawas Surasarang, S.; Keen, J.M.; Huang, S.; Zhang, F.; McGinity, J.W.; Williams, R.O. Hot melt extrusion versus spray drying: Hot melt extrusion degrades albendazole. Drug Dev. Ind. Pharm. 2017, 43, 797–811.
    14. Liu, X.; Lu, M.; Guo, Z.; Huang, L.; Feng, X.; Wu, C. Improving the chemical stability of amorphous solid dispersion with cocrystal technique by hot melt extrusion. Pharm. Res. 2012, 29, 806–817.
    15. Haser, A.; Cao, T.; Lubach, J.; Listro, T.; Acquarulo, L.; Zhang, F. Melt Extrusion vs. Spray Drying: The Effect of Processing Methods on Crystalline Content of Naproxen-Povidone Formulations; Elsevier: Amsterdam, The Netherlands, 2017; Volume 102, ISBN 5124710942.
    16. Repka, M.A.; McGinity, J.W. Influence of Vitamin E TPGS on the properties of hydrophilic films produced by hot-melt extrusion. Int. J. Pharm. 2000, 202, 63–70.
    17. Saerens, L.; Dierickx, L.; Lenain, B.; Vervaet, C.; Remon, J.P.; Beer, T. De Raman spectroscopy for the in-line polymer-drug quantification and solid state characterization during a pharmaceutical hot-melt extrusion process. Eur. J. Pharm. Biopharm. 2011, 77, 158–163.
    18. Saerens, L.; Vervaet, C.; Remon, J.P.; De Beer, T. Visualization and process understanding of material behavior in the extrusion barrel during a hot-melt extrusion process using raman spectroscopy. Anal. Chem. 2013, 85, 5420–5429.
    19. Saerens, L.; Dierickx, L.; Quinten, T.; Adriaensens, P.; Carleer, R.; Vervaet, C.; Remon, J.P.; De Beer, T. In-line NIR spectroscopy for the understanding of polymer-drug interaction during pharmaceutical hot-melt extrusion. Eur. J. Pharm. Biopharm. 2012, 81, 230–237.
    20. Aho, J.; Syrjälä, S. Shear viscosity measurements of polymer melts using injection molding machine with adjustable slit die. Polym. Test. 2011, 30, 595–601.
    21. Dealy, J.M.; Broadhead, T.O. Process rheometers for molten plastics: A survey of existing technology. Polym. Eng. Sci. 1993, 33, 1513–1523.
    22. Ponrajan, A.; Tonner, T.; Okos, M.; Campanella, O.; Narsimhan, G. Comparing inline extrusion viscosity for different operating conditions to offline capillary viscosity measurements. J. Food Process Eng. 2019, e13199.
    23. Chen, Z.L.; Chao, P.Y.; Chiu, S.H. Proposal of an empirical viscosity model for quality control in the polymer extrusion process. Polym. Test. 2003, 22, 601–607.
    24. Yang, Z.; Peng, H.; Wang, W.; Liu, T. Crystallization behavior of poly(ε-caprolactone)/layered double hydroxide nanocomposites. J. Appl. Polym. Sci. 2010, 116, 2658–2667.
    25. Repka, M.A.; Gerding, T.G.; Repka, S.L.; McGinity, J.W. Influence of plasticizers and drugs on the physical-mechanical properties of hydroxypropylcellulose films prepared by hot melt extrusion. Drug Dev. Ind. Pharm. 1999, 25, 625–633.
    26. Low, A.Q.J.; Parmentier, J.; Khong, Y.M.; Chai, C.C.E.; Tun, T.Y.; Berania, J.E.; Liu, X.; Gokhale, R.; Chan, S.Y. Effect of type and ratio of solubilising polymer on characteristics of hot-melt extruded orodispersible films. Int. J. Pharm. 2013, 455, 138–147.
    27. Vo, A.Q.; Feng, X.; Morott, J.T.; Pimparade, M.B.; Tiwari, R.V.; Zhang, F.; Repka, M.A. A novel floating controlled release drug delivery system prepared by hot-melt extrusion. Eur. J. Pharm. Biopharm. 2016, 98, 108–121.
    28. Patil, H.; Feng, X.; Ye, X.; Majumdar, S.; Repka, M.A. Continuous Production of Fenofibrate Solid Lipid Nanoparticles by Hot-Melt Extrusion Technology: A Systematic Study Based on a Quality by Design Approach. AAPS J. 2015, 17, 194–205.
    29. Food and Drug Administration (FDA). Guidance for Industry, PAT-A Framework for Innovative Pharmaceutical Development, Manufacturing and Quality Assurance. In FDA/RPSGB Guidance Workshop; No. September; 2004. Available online: http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/ucm070305.pdf (accessed on 14 October 2020).
    30. Chirkot, T.; Halsey, S.; Swanborough, A. Monitoring the Output of Pharmaceutical Hot Melt Extruders with near Infrared Spectroscopy. NIR News 2014, 25, 15–18.
    31. Whitaker, D.A.; Buchanan, F.; Billham, M.; McAfee, M. A UV-Vis spectroscopic method for monitoring of additive particle properties during polymer compounding. Polym. Test. 2018, 67, 392–398.
    32. Kelly, A.L.; Gough, T.; Isreb, M.; Dhumal, R.; Jones, J.W.; Nicholson, S.; Dennis, A.B.; Paradkar, A. In-process rheometry as a PAT tool for hot melt extrusion. Drug Dev. Ind. Pharm. 2018, 44, 670–676.
    33. Mulrennan, K.; Donovan, J.; Creedon, L.; Rogers, I.; Lyons, J.G.; McAfee, M. A soft sensor for prediction of mechanical properties of extruded PLA sheet using an instrumented slit die and machine learning algorithms. Polym. Test. 2018, 69, 462–469.
    34. Vo, A.Q.; He, H.; Zhang, J.; Martin, S.; Chen, R.; Repka, M.A. Application of FT-NIR Analysis for In-line and Real-Time Monitoring of Pharmaceutical Hot Melt Extrusion: A Technical Note. AAPS PharmSciTech 2018, 19, 3425–3429.
    35. Almeida, J.; Bezerra, M.; Markl, D.; Berghaus, A.; Borman, P.; Schlindwein, W. Development and validation of an in-line API quantification method using AQbD principles based on UV-vis spectroscopy to monitor and optimise continuous hot melt extrusion process. Pharmaceutics 2020, 12, 150.
    36. Guo, X.; Lin, Z.; Wang, Y.; He, Z.; Wang, M.; Jin, G. In-line monitoring the degradation of polypropylene under multiple extrusions based on Raman spectroscopy. Polymers 2019, 11, 1698.
    37. Montano-herrera, L.; Pratt, S.; Arcos-hernandez, M.V.; Halley, P.J.; Lant, P.A.; Werker, A.; Laycock, B. In-line monitoring of thermal degradation of PHA during melt-processing by Near-Infrared spectroscopy. New Biotechnol. 2013, 31, 357–363.
    38. Abeykoon, C.; Martin, P.J.; Li, K.; Kelly, A.L. Dynamic modelling of die melt temperature profile in polymer extrusion: Effects of process settings, screw geometry and material. Appl. Math. Model. 2014, 38, 1224–1236.
    39. Abeykoon, C.; Li, K.; Martin, P.J.; Kelly, A.L. Monitoring and modelling of the effects of process settings and screw geometry on melt pressure generation in polymer extrusion. Int. J. Syst. Control Inf. Process. 2012, 1, 71.
    40. Liu, X.; Li, K.; McAfee, M.; Deng, J. “Soft-sensor” for real-time monitoring of melt viscosity in polymer extrusion process. In Proceedings of the 49th IEEE conference on decision and control (CDC), Atlanta, GA, USA, 15–17 December 2010; pp. 3469–3474.
    41. Tahir, F.; Islam, M.T.; Mack, J.; Robertson, J.; Lovett, D. Process monitoring and fault detection on a hot-melt extrusion process using in-line Raman spectroscopy and a hybrid soft sensor. Comput. Chem. Eng. 2019, 125, 400–414.
    42. Vynckier, A.K.; Dierickx, L.; Voorspoels, J.; Gonnissen, Y.; Remon, J.P.; Vervaet, C. Hot-melt co-extrusion: Requirements, challenges and opportunities for pharmaceutical applications. J. Pharm. Pharmacol. 2014, 66, 167–179.
    43. Stanković, M.; Frijlink, H.W.; Hinrichs, W.L.J. Polymeric formulations for drug release prepared by hot melt extrusion: Application and characterization. Drug Discov. Today 2015, 20, 812–823.
    44. Shah, S.; Maddineni, S.; Lu, J.; Repka, M.A. Melt extrusion with poorly soluble drugs. Int. J. Pharm. 2013, 453, 233–252.
    45. Saerens, L.; Vervaet, C.; Remon, J.P.; De Beer, T. Process monitoring and visualization solutions for hot-melt extrusion: A review. J. Pharm. Pharmacol. 2014, 66, 180–203.
    46. Repka, M.A.; Bandari, S.; Kallakunta, V.R.; Vo, A.Q.; McFall, H.; Pimparade, M.B.; Bhagurkar, A.M. Melt extrusion with poorly soluble drugs—An integrated review. Int. J. Pharm. 2018, 535, 68–85.
    47. Maniruzzaman, M.; Boateng, J.S.; Snowden, M.J.; Douroumis, D. A Review of Hot-Melt Extrusion: Process Technology to Pharmaceutical Products. ISRN Pharm. 2012, 2012, 1–9.
    48. Netchacovitch, L.; Thiry, J.; De Bleye, C.; Chavez, P.F.; Krier, F.; Sacré, P.Y.; Evrard, B.; Hubert, P.; Ziemons, E. Vibrational spectroscopy and microspectroscopy analyzing qualitatively and quantitatively pharmaceutical hot melt extrudates. J. Pharm. Biomed. Anal. 2015, 113, 21–33.
    49. Patil, H.; Tiwari, R.V.; Repka, M.A. Hot-Melt Extrusion: From Theory to Application in Pharmaceutical Formulation. AAPS PharmSciTech 2016, 17, 20–42.
    50. Lang, B.; McGinity, J.W.; Williams, R.O. Hot-melt extrusion-basic principles and pharmaceutical applications. Drug Dev. Ind. Pharm. 2014, 40, 1133–1155.
    51. Maniruzzaman, M. Pharmaceutical applications of hot-melt extrusion: Continuous manufacturing, twin-screw granulations, and 3D printing. Pharmaceutics 2019, 11, 218.
    52. LaFountaine, J.S.; McGinity, J.W.; Williams, R.O. Challenges and Strategies in Thermal Processing of Amorphous Solid Dispersions: A Review. AAPS PharmSciTech 2016, 17, 43–55.
    53. Chokshi, R.; Zia, H. Hot-Melt Extrusion Technique: A review. Iran. J. Pharm. Res. 2011, 5, 1–21.
    54. Jani, R.; Patel, D. Hot melt extrusion: An industrially feasible approach for casting orodispersible film. Asian J. Pharm. Sci. 2014, 10, 292–305.
    55. Kalepu, S.; Nekkanti, V. Insoluble drug delivery strategies: Review of recent advances and business prospects. Acta Pharm. Sin. B 2015, 5, 442–453.
    56. Bhairav, B.A.; Kokane, P.A.; Saudagar, R.B. Hot Melt Extrusion Technique—A Review. Res. J. Sci. Technol. 2016, 8, 155.
    57. Crowley, M.M.; Zhang, F.; Repka, M.A.; Thumma, S.; Upadhye, S.B.; Battu, S.K.; McGinity, J.W.; Martin, C. Pharmaceutical applications of hot-melt extrusion: Part I. Drug Dev. Ind. Pharm. 2007, 33, 909–926.
    58. Breukelaar, A.H.B. Hot Melt Extrusion Technique. WebmedCentral Pharm. Sci. 2011, 2, 135–139.
    59. Schmidt, J.; Marques, M.R.G.; Botti, S.; Marques, M.A.L. Recent advances and applications of machine learning in solid-state materials science. NPJ Comput. Mater. 2019, 5, 83.
    60. Dey, A. Machine Learning Algorithms: A Review. Int. J. Comput. Sci. Inf. Technol. 2016, 7, 1174–1179.
    61. Wosiak, A.; Zamecznik, A.; Niewiadomska-Jarosik, K. Supervised and unsupervised machine learning for improved identification of intrauterine growth restriction types. In Proceedings of the 2016 Federated Conference on Computer Science and Information Systems (FedCSIS), Gdansk, Poland, 11–14 September 2016; Volume 8, pp. 323–329.
    62. Uddin, S.; Khan, A.; Hossain, M.E.; Moni, M.A. Comparing different supervised machine learning algorithms for disease prediction. BMC Med. Inform. Decis. Mak. 2019, 19, 281.
    63. Kotsiantis, S.B.; Zaharakis, I.D.; Pintelas, P.E. Machine learning: A review of classification and combining techniques. Artif. Intell. Rev. 2006, 26, 159–190.
    64. Partheniadis, I.; Toskas, M.; Stavras, F.M.; Menexes, G.; Nikolakakis, I. Impact of hot-melt-extrusion on solid-state properties of pharmaceutical polymers and classification using hierarchical cluster analysis. Processes 2020, 8, 1208.
    65. López del Val, J.A.; Alonso Pérez de Agreda, J.P. Principal components analysis. Aten. Primaria 1993, 12, 333–338.
    66. Hammoudeh, A. A Concise Introduction to Reinforcement Learning; Princess Suamaya University for Technology: Amman, Jordan, 2018.
    67. Morales, E.F.; Zaragoza, J.H. An introduction to reinforcement learning. Decis. Theory Model. Appl. Artif. Intell. Concepts Solut. 2011, 63–80.
    68. Rinnan, Å.; van den Berg, F.; Engelsen, S.B. Review of the most common pre-processing techniques for near-infrared spectra. TrAC Trends Anal. Chem. 2009, 28, 1201–1222.
    69. Xu, L.; Zhou, Y.P.; Tang, L.J.; Wu, H.L.; Jiang, J.H.; Shen, G.L.; Yu, R.Q. Ensemble preprocessing of near-infrared (NIR) spectra for multivariate calibration. Anal. Chim. Acta 2008, 616, 138–143.
    70. Almeida, A.; Saerens, L.; De Beer, T.; Remon, J.P.; Vervaet, C. Upscaling and in-line process monitoring via spectroscopic techniques of ethylene vinyl acetate hot-melt extruded formulations. Int. J. Pharm. 2012, 439, 223–229.
    71. Zeaiter, M.; Roger, J.M.; Bellon-Maurel, V. Robustness of models developed by multivariate calibration. Part II: The influence of pre-processing methods. TrAC Trends Anal. Chem. 2005, 24, 437–445.
    72. Bi, Y.; Yuan, K.; Xiao, W.; Wu, J.; Shi, C.; Xia, J.; Chu, G.; Zhang, G.; Zhou, G. A local pre-processing method for near-infrared spectra, combined with spectral segmentation and standard normal variate transformation. Anal. Chim. Acta 2016, 909, 30–40.
    73. De Maesschalck, R.; Estienne, F.; Verdú-Andrés, J.; Candolfi, A.; Centner, V.; Despagne, F.; Jouan-Rimbaud, D.; Walczak, B.; Massart, D.; De Jong, S. The development of calibration models for spectroscopic data using principal component regression. Internet J. Chem. 1999, 2, 1.
    74. Blazhko, U.; Shapaval, V.; Kovalev, V.; Kohler, A. Comparison of augmentation and pre-processing for deep learning and chemometric classification of infrared spectra. Chemom. Intell. Lab. Syst. 2021, 215, 104367.
    75. Fearn, T.; Riccioli, C.; Garrido-Varo, A.; Guerrero-Ginel, J.E. On the geometry of SNV and MSC. Chemom. Intell. Lab. Syst. 2009, 96, 22–26.