2022 Volume 10 Pages 257-287
Coffee attracts consumers worldwide for its unique sensorial properties. Its unique flavor is affected by numerous factors. The biochemical properties associated with geographical features are among the essential aspects that may modulate coffee’s distinct sensorial profiles, and may be employed for its origin determination and authentication. This systematic literature review served to assess the newest techniques for coffee authentication, origin determination, and adulteration detection from analytical and nondestructive approaches. This study focused on the last 10 years’ high-quality research in the field. Accordingly, 78 articles using both analytical and nondestructive methods for the determination of coffee origin and fraud detection were found in leading journal databases. Apart from the compound profile and instrumentation, data analyses including statistics, machine learning, and multivariate models which have been commonly used are discussed as well. In addition, other important information, such as data validation methods and the predictive capability of the above techniques are also reflected in this review.
Coffee is one of the most popular beverages and one of the most traded commodities worldwide. Green coffee beans can travel thousands of miles from their country of origin to consumer markets before eventually reaching the table. Widely grown across the tropics, coffee is an important export crop for many developing countries and a significant contribution to the livelihoods of local farmers in said countries. Brazil, Vietnam, Columbia, and Indonesia account for the vast majority of global production, whereas the European Union, USA, and Japan are the world’s biggest importers [1]. Generally, coffee consumers have a preferred or favorite coffee origin. This consumer choice is related to sensory properties, such as aroma and mouthfeel. Coffee has a unique organoleptic profile associated with its growth geographical location. Single-origin coffee refers to coffee cultivated in a specific microclimate and typically sourced from a certain geographical place, such as a farm or multiple farms, or plantations, or a region within the same country. This type of consumption has been increasing all over the globe [2].
However, high-quality single-origin coffee is prone to misleading labels, false declarations, and fraudulent practices to increase their economic profit. Recently, the Federal Food Safety and Veterinary Office of Switzerland reported a falsely declared “100% Arabica” coffee substituted by the cheaper Robusta coffee [3].
Coffee fraud may imply, for instance, counterfeiting high-quality and specialty coffee beans with lower quality or defective beans. Another possibility is the falsification of geographical origin information [4]. As an indicator of product and process quality, product origin shows increasing importance for the business and for informing consumers’ purchasing decisions.
In response to consumer demands for authenticity of coffee origin, various strategies encompassing a broad range of technology and scientific techniques have been applied to assure this point. In the past decade, studies pertaining to coffee origin authentication, determination and classification were done utilizing near-infrared (NIR) [5, 6, 7], Fourier Transform Mid-infrared (FT-MIR) [8], Terahertz Spectroscopy [9], e-nose or e-tongue sensors [10], [11], and UV-visible spectroscopy [12]. Moreover, several studies with similar purposes were conducted by gas [14, 15] or liquid chromatography [16, 17, 18] coupled to mass spectrometry, nuclear magnetic resonance (NMR) [19], [20], and polymerase chain reaction (PCR) [21, 22, 23].
These techniques can be categorized into nondestructive and analytical approaches. The nondestructive technologies allow a rapid analysis and are less laborious, considerably saving costs, and require little or no disruption of the biochemical potency of the sample [24]. On the other hand, the advantage of using an analytical approach in the geographical determination of coffee is associated with the possibility of analyzing important markers indicative of its origin [25].
Nondestructive equipment produces large number of spectral signals that can be count as variables, therefore it is attractive in their data processing and chemometric analysis. Machine learning and deep learning algorithms are often used to interpret the data. Nevertheless, the data processing of analytical techniques can be more advanced when coupled to the bioinformatics aspect to reveal the important compound markers. Nondestructive methods require a large samples dataset to build the initial model. In addition, data interpretation can be difficult. Therefore, other studies use analytical approaches, such as chromatography and mass spectrometry, or a combination.
Regarding the capability of the analytical method to identify coffee markers, the sensorial properties of coffee heavily rely on its biochemical compound content. Its unique taste is affected by numerous factors. From the very beginning is the environment where it is planted, followed by the coffee cherry growth stage, the local farmers’ postharvest tradition and then successively up to the method of brewing the coffee for serving in the cup.
Among the factors influencing the hedonic preference of coffee consumers, aroma and mouthfeel are by far the most important. The aroma is associated with volatile compounds, while the mouthfeel is generally based on its lipid components. Both compounds can be analyzed using a metabolomics approach. According to Fiehn [26], metabolomics is the comprehensive analysis of global metabolites of a biological system. Similarly, the comprehensive study of lipid compounds is commonly called Lipidomics [27]. Lipidomics is a developing research field supported by the improvement of various analytical methods, especially mass spectrometry and bioinformatics [28].
Most studies applying chromatography and mass spectrometry aim to explore the metabolite profile and identify the key flavors of coffee identifying different origin locations. Furthermore, numerous studies show that each coffee origin is characterized by distinct biochemical compounds.
This review aimed to investigate the most recent techniques for coffee authentication, origin determination, and adulteration detection from both analytical and nondestructive approaches. Despite the existence of previous similar reviews, this is the first systematic literature review (SLR) following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [29].
The objective of this SLR is to offer a thorough description of the latest findings in the latest updated research context and the scope for research questions for future study. This study aimed to deliver an accurate scientific report and avoid bias; therefore, this study adopts SLR methodology by Moher et al. [29]. Figure 1 presents the study flow, which began by identifying studies in literature databases according to certain search words, followed by several screening steps. The article title was the first screened, followed by the abstract. Finally, the full text was rigorously studied to be included in the primary literature.
The PICO (Population, Intervention, Comparison, and Outcome) framework was used to define the inclusion criteria (Table 1). The PICO framework is a model for conducting a reference search that divides a formulated research question into four distinct components: the population of interest, the applied intervention, the comparison or controls, and the measured outcome.
Figure 1: PRISMA flow chart for primary literature screening.
Articles were obtained from well-established databases, i.e., Scopus, Springer, ScienceDirect, PubMed, and Wiley-Blackwell. Each database uses a different style of search syntax and operators. The keywords or syntax used are indicated in Table 2. The search strategy yielded several numbers of article (n), which is indicated in Figure 1. The syntax limited the search for peer-reviewed publications (original and review papers), in English language, journal articles, and published within the past 10 years, from 2012 onwards. These records were exported to Mendeley reference manager (Ver 1.19.8), following the removal of duplicated studies and irrelevant article types.
| Framework | Criteria |
|---|---|
| Population (P) | All coffee varieties in the Coffea (genus) |
| Intervention (I) | Geographical authentication, origin determination |
| Comparison (C) | Destructive vs. Nondestructive approaches |
| Outcomes (O) | Comparison of both approaches |
Various data were extracted from the final list of included studies, namely, the publication year, techniques or instrumental approach used, country of origin of coffee samples, classification model algorithm and performance evaluation, feature selection, associated information on coffee processing, and key findings.
| Database | Keywords and syntax | |
|---|---|---|
| ScienceDirect | : | [(coffee OR coffea) AND (geographic OR origin OR region OR country) AND (authentication OR determination OR discrimination)] year 2012–2022 |
| Scopus | : | TITLE-ABS-KEY [(coffee OR coffea) AND (geographic* OR region*) AND (authenticat* OR origin OR provenance) AND (determin* OR discriminat*)) AND PUBYEAR > 2011 AND (LIMIT-TO (PUBSTAGE, “final”)) AND (LIMIT-TO (DOCTYPE, “ar”) OR LIMIT-TO (DOCTYPE, “cp”) OR LIMIT-TO (DOCTYPE, “ch”) OR LIMIT-TO (DOCTYPE, “re”)) AND (LIMIT-TO (LANGUAGE, “English” )] |
| Springer | : | with all of the words: coffee geographic* origin country with at least one of the words: authenticat* discriminat* determin* classif* where the title contains: coffee year: 2012–2022 |
| PubMed | : | “coffee OR coffea” in Title and “(origin OR region* OR geographic* OR country) AND (trac* AND authenticat* OR determin* OR discriminat* OR classification)” in Abstract |
| Wiley-Blackwell | : | “coffee OR coffea” in Title and “(origin OR region* OR geographic* OR country) AND (trac* AND authenticat* OR determin* OR discriminat* OR classification)” in Abstract Year 2012–2022 |
Screening results using the PRISMA approach are indicated in Figure 1. The search strategy identified a total of 642 records. Then, each record was rigorously assessed for eligibility, after which 78 original research papers were retrieved to be reviewed in detail. The number of studies using analytical and nondestructive methods was 55 and 23, respectively. In addition to the original papers, seven review articles on similar topics to the current study are listed and briefly discussed in Table 3. Generally, said reviews discussed and identified common methodologies to determine coffee’s geographical origin, adulteration, and fraudulent practices. However, none of the articles performed a systematic review. In contrast, the present study performed a more detailed and structured assessment of the most updated research in coffee origin classification, determination, and authentication. The importance, advantages, and features of this review compared to existing reviews are shown in Table 3.
3.1. Analytical approachesNumerous studies on biochemical coffee profiling based on broad range metabolome analysis have been conducted. The key information of the 55 studies on coffee origin determination, authentication, and adulteration using the analytical approaches are exhaustively summarized and listed according to publication date in Table 4. Early studies on coffee classification used Nuclear Magnetic Resonance (NMR)-based fingerprinting and elemental analysis using Inductively Coupled Plasma Mass Spectrometry (ICP-MS)-based. Using NMR approaches, coffee green beans from different countries [30] and roasted coffee from several continents were classified [31], as well as quantification of adulteration of coffee varieties (Arabica and Robusta) [32]. In addition, Arana et al. [33] employed NMR to distinguish Colombian coffee from that from other origins. These studies found that the NMR spectra of coffee samples showed significant resonance from caffeine, sugar compounds, chlorogenic acids, fatty acids, and amino acids. In addition, recent research using the NMR approach also found that lipids, acetic acid, lactic acid, and quinine were discriminative compounds for several Indonesian coffees [19, 34].
ICP is often used for fingerprinting the elemental compounds and or isotope ratios of coffee samples. Green and roasted coffee beans analyzed by ICP-MS and ICP-Emission Spectroscopy showed negligible differences in the elemental composition. Furthermore, harvest year and degree of ripeness were nonsignificant [35]. Another study employing ICP-optical emission spectrometry (OES) found that metal element content could discriminate the coffee origin from different countries in South America [36], the inter-Mexican region [37], cross-continental samples [38], Ethiopian coffee from 11 different regions [39] and different postharvest process [40], and Jamaican coffee against non-Jamaican [41]. In agreement with Valentin and Watling [35], Habte et al. [39] confirmed that harvest year did not significantly influence coffee’s elemental compounds.
The more recent studies in Table 4 use chromatography combined with mass spectrometry. Gas Chromatography-Mass Spectrometry (GC/MS) volatile compound profiling served to determine the coffee origin among several countries on different continents; this study used postharvest process as a classification variable [42], and also civet coffee (kopi luwak) discrimination against non-luwak coffee from the Philippines [43]. Other volatile metabolomic approaches were applied for authentication by analyzing variation in its roasting levels [44, 45]. Overall, volatile profiling found that pyrazines, furans, and other aromatic hydrocarbons influenced the coffee origin classification. GC/MS untargeted metabolomics profiling was used to determine coffee from various places in Indonesia [16], finding that the metabolome profile of green Arabica coffee beans differed from Robusta beans, as well as, the differentiation of roasted coffee beans from various island. Recently, the same researchers indicated that the postharvest process is the most discriminative aspect in coffee, followed by geographical origin [46].
The volatile compounds of coffee from 7 different cultivars in Hainan (China) were profiled using headspace solid-phase microextraction (SPME) GC/MS, with unsatisfactory results for the differentiation of green coffee beans. Yet, combining several analysis including fatty acids, amino acids, and proteins, the study could successfully classify the Robusta Hainan coffee samples [13]. Volatile profiling was not effective in discriminating the origin of raw green coffee because of the lack of aroma at this stage. However, several studies could discriminate raw green coffee beans. For instance, a study using Ultra Performance Liquid Chromatography-Mass Spectrometry (UPLC-MS) to determine phenolic [18] and alkaloid compound [47] profiles could classify Ethiopian coffee (east, northwest, west, and south regions). As well as Yemeni [99] and Ethiopian [49], coffee green beans were classified using elemental analysis by ICP–OES.
Other authentication method employed various techniques. For instance, photon activation analysis (PAA) using a radioanalytical method in the elemental analysis was employed for classifying three South American coffee beans and for distinguishing washed and natural process coffee [50]. Proton-Transfer-Reaction (PTR-ToF-MS) showed different volatile compound profiles in coffee from Brazil, Ethiopia, and Guatemala [51]. High-Resolution Continuum Source Atomic Absorption Spectrometry could classify the espresso extracted coffee based on its elemental profile [52]. Other studies applied direct-injection electrospray-MS for fingerprinting and low-temperature plasma ionization-MS for rapid analysis [53]. The carbon isotope ratio (δ13C) of caffeine and that of the whole volatile fraction have been analyzed using GC-Carbon Isotope Ratio-MS [54]. Laser-Induced Breakdown Spectroscopy (LIBS), a new technique to detect and quantify coffee adulterants (chickpeas, maize, and wheat), could identify <0.6% adulterations in coffee [55].
A few studies used a DNA-based approach, Polymerase Chain Reaction (PCR)−Denaturing Gradient Gel Electrophoresis (DGGE) to understand the microbial community existing in coffee from different origins and processing. The study found that geographical origin has little effect on microbial diversity [21]. In addition, Ferreira et al. [22] used DNA markers of adulterants such as corn, barley, and rice to quantify the percentage of noncoffee contents by Real-time PCR. Combes et al. [23] employed high-resolution melting (HRM)–PCR to identify adulterated coffee in both green and roasted beans, resulting in a 1% threshold for adulterants content detection. Recently, the HRM method was applied to brewed Thailand samples, showing promising results for the detection of Arabica–Robusta admixtures in brewed coffee [57].
Interestingly, a couple of studies performed untargeted fingerprinting analysis using HPLC without MS. High Performance Liquid Chromatography (HPLC)-UV fingerprinting was used to classify coffee samples from several countries and continents with varied roasting levels [58]. The same group used HPLC-Fluorescence Detection (FLD) to identify an admixture of coffee from different origins [59], and classify coffee origin based on countries, variety (Arabica and Robusta), and roasting degree [60]. HPLC-FLD achieved a richer chromatogram fingerprint than HPLC-UV.
With respect to analytical approaches, several studies used specific chemical compounds such as antioxidant compounds [61], total phenolic compounds, protein, and total lipids [62] as variables for coffee origin classification. However, these studies did not find a significant origin classification when data were modeled with multivariate analysis, as confirmed by Alnsour et al. [63].
Generally, various analytical methods used to determine the geographical origin of coffee from numerous countries, establish authentication methods, and detect adulteration with noncoffee materials or addition of lower value substances. Overall, NMR is a high-throughput analytical device. This method requires little sample preparation, and can separate substances based on their NMR fingerprint. In comparison to MS-based approaches, NMR has limited sensitivity. Given its effectiveness and superior separation capabilities, numerous coffee research uses chromatography coupled to MS which has been proven for its reproducibility.
Moreover, high-resolution MS offers accurate mass measurements and can lead to the prediction of empirical formulas for unidentified compounds. This tool is widely employed when coupled with GC, and supported by fragmentation libraries (commercial and open-source) that assist metabolite identification. However, the method requires chemical derivatization and cannot be used for larger, nonvolatile substances. New HPLC methods have increased its separation efficiency, and together with MS permits the identification of substances without chemical derivatization. Additionally, the automated sampling facilitates the daily assessment of several samples. In terms of data comparison, it is difficult because of uniform ionization and fragmentation. Further discussion on data analysis and classification/prediction models will be discussed in section 3.3.
The 23 selected articles on this topic are shown in Table 5. During the last 10 years, numerous nondestructive approaches for coffee origin determination, authentication, and variety assessment of the raw green, roasted, and brewed coffee samples have been extensively done. Most of the studies used spectroscopy-based methods. Spectroscopy is a fast-growing technique due to its speed, simplicity, safety, and ability to examine several characteristics simultaneously without requiring lengthy sample preparation[64]. Particularly, spectroscopic procedures in the visible, near, and midinfrared regions are rapid, almost chemical-free, inexpensive, and sample-processing-free techniques widely used to predict the chemical composition of coffee, making them suitable for routine application.
Several studies on this topic use single equipment, but most applied combined approaches to achieve their goals. Infrared (IR)-based technology is the most employed method in nondestructive analysis. It is categorized into three regions, i.e., near IR (NIR) from 0.77 to 2.5 µm (corresponds to wavenumber = 13000–4000 cm−1); mid-IR(MIR) from 2.5 to 15 µm (4000–400 cm−1), and far IR >25 µm (<400 cm−1) [65]. This technique can create a spectral “fingerprint” of coffee samples by direct measurement. However, for coffee authentication or origin determination studies, only a certain range of wavelengths is meaningful. For instance, a study using FTIR coupled to a specific detector selected wavelengths ranging between 600 and 1000 cm−1 for differentiation of two cultivation systems of roasted coffee (organic and conventional) [66], while other studies using MIR employed spectral features between 2970 and 3600 cm−1 to classify Robusta and Arabica coffee [67].
The studies listed in Table 5 used NIR for numerous purposes, e.g., differentiation of modern and traditional coffee cultivars from Brazil [5], regional classification of Brazilian coffee samples [115], origin determination of coffee from Cuba, Ethiopia, Indonesia (Bali, Java, and Sumatra), Tanzania, and Yemen [8] comparison between South American and Asian coffee [68], and to detect impurities (Corn, Rice, Barley, Soybeans, Coffee husks) in Arabica roasted coffee samples as well as to separate South and Central American coffee samples [69]. All the above studies used >100 total samples. The number of samples is probably one of the good parameters for obtaining robust classification in NIR approaches.
Other noninvasive methods listed in our study are the combination of several tools. For instance, a study compared the effectiveness of Attenuated Total Reflectance Mid-Infrared (ATR-MIR), NIR, and 1H-NMR. ATR-MIR led to better classification of coffee species and country of origin than NIR and 1H-NMR to distinguish Colombian coffee from counterfeit beans [70].
Furthermore, voltametric sensor technology, particularly the electronic nose (E-nose) and electronic tongue (ET), was used in several studies. Some studies used the sensor together with other analytical approaches to confirm sensor data [48]. For example, the E-nose and GC/MS were used to determine the origin of civet coffee (the Philippines) against regular coffee from the same region [71] and to discriminate against coffee from Brazil, Ethiopia, Guatemala, Costa Rica, and Peru [72]. As well as E-nose and GC/FID which was used in classifying coffee samples from several countries [73]. The E-nose and GC were used to assess the volatile aroma compounds of coffee samples. Meanwhile, several ET sensors were used to distinguish samples from different origins and cultivation practices (organic vs. conventional) as well as the altitude of Mexican coffee plantations [11]. On this subtopic, one study in Table 5 used a UV-Vis spectrometer to distinguish Indonesian civet coffee (kopi luwak Lampung) from adulterated kopi luwak [12]. Recently, Terahertz spectroscopy was used to discriminate coffee samples from Kenya, Kilimanjaro, and Yunnan (China). One of the samples was roasted at three different levels (light, medium, and dark). The result of model classification was satisfactory [9]
Overall, coffee origin determination, authentication against contaminated samples, and in-country origin determination can be performed by simple and rapid, nondestructive approaches. Most of the studies used a lot of samples with multiple replications. Furthermore, the spectral data from these approaches were processed with multivariate analysis or employing machine learning modeling.
3.3. Multivariate model and data analysisPrior to data analysis, several aspects of data collection are essential to obtain robust classification and coffee origin determination or authentication. The number of samples and representation of biological samples from a certain geographical origin follows a rule of thumb, especially when applying a multivariate analysis or machine learning model. Using highly representative samples for creating a calibration model will be more accurate in predicting the studied class. Therefore, this review indicates not only the number of samples used but also the model performance evaluation method.
Comparing the classification model performance among studies is difficult due to technical bias, such as HPLC or GC/MS methodology or NIR setup. A study using both analytical and nondestructive approaches on the same samples is required for a valuable comparison. A study by Monteiro et al. [74] (Table 4) may illustrate this comparison, where volatile organic compounds profiling (based on PTR-MS) and nondestructive (based on NIR) was used to classify organic vs. conventional coffee farming. The data was processed with several machine learning models using cross validation, followed by validation by an external dataset. In the model both method can be compared for the percentage of accuracy.
Spectral data from analytical platforms like NIR, MIR, NMR, chromatography, and MS, contain molecular information that can serve as fingerprints for coffee origin, species, and types. So that spectral data can be useful for coffee authentication, statistical analysis are frequently required to reduce data dimensionality, such as the identification of spectrum regions relevant to quality parameters, pattern recognition, and detecting outliers. Unsupervised exploratory approaches, including PCA, factorial analysis, Soft Independent Modeling of Class Analogy (SIMCA), and cluster analysis, are frequently employed for this purpose. The main aim of unsupervised modeling is to explore the natural sample grouping [75]. This model is not aimed at finding important variables in large data, such as in metabolomics. However, numerous studies used this model and arbitrarily selected the important compounds by observing the loading plot.
Other studies used better data modeling, such as supervised machine learning. Several algorithms can provide insights into feature selection, such as Linear Discriminant Analysis (LDA) [18, 36, 76], Partial Least Square (PLS)-Discriminant Analysis (DA) [51, 77], PLS Regression [58], Orthogonal PLS-DA [78, 29, 32], Random Forest (RF) [53], support vector machine (SVM) [61, 74, 79], and k-nearest neighbour (k-NN) [74, 79].
The model used in all studies is indicated in Tables 4 and 5. These algorithms have been adopted to handle metabolomic datasets in a supervised method. In the supervised model, the class of observation is already decided for the classification of the sample data. Therefore, the subjectivity of the human interference in the mathematical formulation can be considered not the best fit for the grouping from random set of data, yet the result of the model might be as satisfactory [80]. Moreover, model resulted in the DA can be more stable [81].
Another important part of data analysis is cross validation (CV). Several studies used CV to reduce bias and avoid overfitting (Tables 4 and 5), whereas others split data into calibration/training datasets and validation/test samples [104]. Furthermore, normally a DA uses root mean square error to indicate the error level of a model, as shown in Amalia et al. [46]. Yet, other models prefer to use R2X and Q2Y to evaluate the goodness of fit of their model [34]. Important features for discrimination are normally selected based on the model algorithm capacity. For example, the PLS family model uses a Variables Importance in Projection (VIP) score or S-plot to indicate the influence of different variables in the classification [109]. An exhaustive list of models and evaluation parameters, as well as the feature selection methods can be found in Tables 4 and 5.
Finally, a general comparison of the advantages and disadvantages of analytical and nondestructive approaches is illustrated in Table 6. Furthermore, future research ideas on coffee origin classification and/or authentication are indicated in Table 7.
| Method | Advantages | Limitations |
|---|---|---|
| Analytical (in general) | ・Proper technique for elucidation of the biochemical fingerprint of food materials. ・Golden standard to find the information of important biochemical compounds which drive sample discrimination. ・In certain approaches, it can be interpreted immediately real time, employing simple statistical analysis. |
・Labor intensive since it requires trained operator to comprehend the advanced technology. ・High investment for purchasing the whole system and certain components. ・Requires complex sample preparation and pretreatments. |
| Nondestructive (in general) | ・Minimum sample preparation (e.g., does not require extraction, derivatization, and other pretreatments). ・Relatively simple and rapid operation. ・May accurately be implemented when the training model have been developed, and ・In certain case, can be used along with the different production steps to provide real-time information. |
・Could not be used to discover the important biomarker of a product. ・Need a large number of samples to create an accurate determination. ・Could not be interpreted immediately to indicate the discrimination factor since the spectral patter might be similar, and thus it should be proceeded to the chemometrics data analysis to find the discriminative wavelength. |
| No | Ideas for future prospective studies |
|---|---|
| 1 | Authentication of coffee based on untargeted metabolomics analysis have not been studied comprehensively especially for the samples involving the variability of origins coupled to the variations of degree of roasting as well as the different harvest seasons. |
| 2 | Volatile compounds and lipid fraction investigation in association with geographical origin of coffee with variation on their altitudes and postharvest processing. |
| 3 | As a future direction, LC/MS represents a potential tool for targeting nonvolatile metabolites in brews, including primary and secondary metabolites. |
| 4 | Untargeted HPLC-UV or HPLC-FLD have not been thoroughly studied, specifically for the coffee green beans and roasted bean of the highly valued coffee such as kopi luwak and Monsoonal processed coffee. |
| 5 | Numerous studies identified potential markers compounds for geographical origin determination. However, the confirmation of the selected markers after long storage has never been found in the literature. |
Overall, the findings of this review can be beneficial for both the coffee industry and scientific community. Both analytical and nondestructive methods are relevant. The classification of coffee expanded over time, earlier studies still classify and identify coffee types (Arabica–Robusta), yet currently numerous studies were done employing diverse origins of coffee. However, the in-country geographical classification remains an issue since coffee within nearby cultivation areas cannot be separated. Most studies did not consider the effects of roasting levels, postharvest process, and storage of green and roasted beans.
