Applied Science and Engineering Journal for Advanced Research

1. Introduction

Due to the globalization of economy and free trade agreements (FTAs) among countries, all industrial sectors including agricultural products sector cope with unprecedented steep competition due to reduced trade barriers and simplified customs procedures. These globalization trends also present new opportunities by allowing business entities to target a much larger group of consumers beyond geographical boundaries in the globally connected marketplace.Due to the limited information of products from remote regions and even foreign countries, however, consumers in the perishable food sectors such as meat and poultry eager to verify the adulteration and authenticity in real time (Kamruzzaman et al., 2015). To this end, almost all member countries of the Organization for Economic Co-operation and Development (OECD) implement digital traceability systems to ensure food safety and promote transparency throughout agricultural food supply chains. For a comprehensive review of digital traceability systems of OECD countries, the readers are strongly recommended to refer to the study (Charlebois et al., 2024) and cited studies there.

In this paper, we focus on the traceability and transparency in digital food supply chains in South Korea. In particular, we are interested in beef supply chain, tracking the movement of beef from the location of farms and farmers to the location of finished beef. In Korea, the South Korean Ministry of Agriculture, Food and Rural Affairs and the Ministry of Science and ICT started to collect data such as cattle rearing, slaughtering, packing and where is to be sold in January 2019 (Euro Meat News, 2023). In particular, the South Korean authorities intended to collect and store all data automatically through attached Bluetooth devices to the cattle and maintain data integrity by storing all credential information on the blockchain. Ultimately, such digital traceability system of beef supply chain would enable the authorities to detect any beef hygiene and safety issues more quickly and react as soon as possible (Charlebois et al., 2024).

In addition to effort from the South Korean government, several major companies in Korea also collaborated together for the purpose of enhancing food traceability. For instance, the Korea Telecom (KT) and Nongshim Data System (NDS) worked

For our study, we decided to take only the first batch of data sets over three years to control this preliminary study under manageable size. The finalized data contains a total of 514,020 records along with unique identification number, cattle biological characteristics (Birth Date, Breed Type, and Gender), farm information (Farms Location and Owner), meat processing characteristics (Slaughter House and Date, Months and Weight when Slaughtered), and Meat Grade. We present Table 1 for the detailed list of input variables with data types.

Table 1: List of Variables

Few variables warrant further explanation. We first explain three breeds of cattle briefly. In Korea, Hanwoo is a Korean native species of cattle with a brown hair and is most closely related to two taurine breeds, theHolsteinand theJapanese Black, but separate from the breeds of China and Europe (Kim et al., 2010; Lee et al., 2014). With the explosive growth of theSouth Koreaneconomy since 1960s, many researchers have engineered this breed to produce high quality of meat (Lim et al., 2016) and, as a result, its unique meat trait is mostly different from other breed meats in terms of pleasant savory and chewy taste. Therefore, in Korean beef market, Hanwoo meat is most preferred to cheaper imported beef and is considered the highest quality by many Koreans even with its highest price. Further, many Koreans consider Hanwoo meat a cultural icon for a special-day due to its best quality in the world and Hanwoo is most desired ingredient of Korean traditional foods for holiday dishes.

Next, meat (or beef) cattle are cattle raised mainly for meat. In particular, meat from cows who have never produced calves, from hybrids of Hanwoo and foreign species, or cow hydrogen is regarded as beef cattle meat.

Our findings will provide insights on logistical relationships between farms and slaughter houses across cities, regions, and national boundaries.

Finally, we explore how accurately a machine learning model (e.g., decision tree (DT) classifier) can predict the quality of beef grade given limited information of cattle along with supply chain information. Currently, the South Korean government issues beef quality certificate based on the level of marbling, meat color, fat color, and meat texture determined by trained professionals. Therefore, the proposed prediction model of beef quality will provide supplemental information to government issued certificate that enable consumers to make informed choices by simply referring to the limited information of cattle on the product label.

4. Multi-Factor Analysis on Meat Grade

4.1 Measuring Impact of Biological Factors on Meat Grade

In this section, we intend to consider several biological factors to identify the direction and magnitude of their impact on the meat grade. To this end, we converted the original categorical meat grade (others, 3, 2, 1, 1+, and 1++) to a numerical scoring grade (0, 1, 2, 3, 4, and 5) so that we can compute average values of meat grade for a chosen dimension such as cattle breed or gender. To begin with, we first conducted a preliminary analysis to verify a well-known relationship between cattle breed and meat grade. In our data sets, there are three cattle types (Hanwoo, beef cattle and milk cow), and, among them, Hanwoo occupies the largest portion (90.5%) of data sets followed by beef cattle (6.1%) and milk cow (3.4%). Our data analysis confirms that Hanwoo produces the highest meat grade on followed by beef cattle and milk cow (2.86 vs. 1.45 vs. 0.53) and the difference of meat grade among cattle types is statistically significant (t-test, p < 0.01). We also noted that within the same cattle breed, meat quality values range significantly, implying that there are other factors that affect the meat quality.

The next factor we considered is gender. In our data sets, the proportion of neutralized male, female, and male is 49.1%, 49.9%, and 1%, respectively, and hence the majority of records are from neutralized male and female.

It is true even for milk cow type that is raised mainly for milk, and hence it produces only low grade meat about 1.0. It is much clearer to see a positive relationship between weight and meat grade in meat cattle type in both female and neutralized male gender as long as they are in ranges of similar weights. We did not show the relationships for male gender of all cattle types here mainly because of its minor percentage in terms of records.

Figure 2: Average Meat grade with Gender, Breed Type, and Weight

It, however, presents slightly different impact patterns of weight on meat grade of Hanwoo. For female Hanwoo, its meat grade improves from 1 to 3.4 as its weight changes from 170kg to 470kg, following a linear relationship in general. Interestingly, its meat grade improves at an increasing rate in the interval of weights less than or equal to 400kg, but then improves at a decreasing rate thereafter. Notably, its meat grade does not improve at all but rather fluctuates significantly after its weight is greater than 450kg. In contrast, we noticed some different patterns from Hanwoo meat from neutralized male. First, the meat grade of neutralized male Hanwoo is superior to that of female Hanwoo (between 1.5 and 4.5 vs. 1.0 and 3.4) partly because neutralized male Hanwoo weighs between 228kg and 575kg and hence has a better chance to fit into an optimal weight range (450kg and 590kg) that produces the highest quality of meat than female Hanwoo that weighs between 168kg and 473kg. In addition, Hanwoo meat from neutralized male is steadily increasing at a decreasing rate as its weight increases without any acceleration and deceleration observed from female Hanwoo.

4.2 Measuring Relationships between Farm Cities and Meat Grade

One notable finding we like to address is a possible conflict between findings from our analysis and commonly accepted fact about the best places to taste the best quality of meat in Korea. For example, it is well known that Hongchun (5,437 c and 2.741 g) and Hoengseong (1,314 c and 1.888 g; it is not shown explicitly in Figure 3 but is located just below Hongchun city) in the north-east region (i.e., Gangwon province) are one of the best places to taste the best meat in Korea. This finding surprised us because Hoengseong is best known for its Hanwoo cattle and the county began a strategic marketing campaign to brand itself as the origin of the highest quality beef in Korea along with environment well-suited for cattle farming. A simple Google query for “best places for Hanwoo” endorses it by returning these two cities and three other cities such as Jangheung (12,106 c and 3.447 g), Jeongeub (11,866 c and 3.161 g), and Andong (10,251 c and 3.092 g) as query outcomes. However, unlike three other cities, Hongchun and Hoengseong are not validated from our analysis. That is, their average meat grades are much lower than those of other recommended cities based on our analysis and Google query. While this conflict warrants further investigation in future studies, one of few explanations includes the possibility of having biased and limited data sets. In addition, this conflict does not exclude the possibility that few farms in these two cities may produce meat with the highest grade although average meat grade of all farms in these cities is relatively low.

4.3 Identifying Slaughter Houses with High Meat Grade

In this section, we like to identify and visualize (slaughter) houses that tend to produce higher meat grade than in other houses. To this end, we first computed averages of meat grade for each slaughter house by breed types and visualized them in column charts in Figure 4. To simplify charts, we only showed top 11 houses by removing six houses that process less than or equal to 1,000 cattle. In these figures, left and right y-axis represents the number of cattle processed and the average meat grade from each house. In particular, in Figure 4, the data label in the column chart segment with blue color represents the number of Hanwoo processed in each house, while numbers of other breed types are not shown to simplify the chart.

Figure 5: Processed Cattles and Meat Quality by Gender in Slaughter Houses

According to Figure 5, the 3rd house processes a higher proportion of neutralized male cattle than the 6th house does, its average meat grade is lower than that of the 6th house (2.952 g vs. 3.112 g). In addition, when we compare average meat grades from few houses (i.e., the 5th, 10th, and 11th) that processed a similar proportion of neutralized gender cattle, their meat grade are also lower than that of the 6th house. Considering these observations, we conclude that the reason of high meat grade of the 6th house does not come from imbalanced gender or breed type but from high quality cattle available from farms near this house. Note that the 6th house is located in the mid-west region where two cities Hongseung (3.100 g) and Sejong (3.134 g) that produce high quality meat are located. We also note that other houses (e.g., the 3rd (2.952 g), 4th (2.919 g) and 5th houses (2.8111 g)) with high grade meat are also located nearby cities like Jangheung, Jeongeub, Andong and Youngju.

5. Beef Quality Classification Model

While our main objective of this study is to exam the significance of multiple factors on meat grade and identify and visualize various supply chains between all farms and all slaughters in Korea, we also like to calibrate a machine learning model (e.g., decision tree (DT) to identify cattle that are most likely to produce the highest-grade meat. Note that DT is a non-parametric algorithm adopted by many researchers in computer sciences and engineering because of its white box characteristic,

representing top x% of population based on the estimated probability of being cattle with the highest-grade meat. By definition, a lift chart for a random model is always a diagonal line because it will statistically correctly identify x% of positive records (= meat records with highest-grade) when top x% of population are classified. The lift chart of an imaginary ideal model will correctly identify all positive records using only the per cent of positive records in the population. We presented the lift charts in Figure 7.

Figure 7: Lift Charts of DT, Random, and Ideal Model

In Figure 7, a green line represents the lift chart of DT model while a red and a blue line represents the lift chart of an ideal and a random model, respectively. According to Figure 7, an ideal model required about 9% of samples to identify all positive samples and hence when top 20% of cattle are chosen for a prediction task, it identified 100% of positive samples. In contrast, a random model identified only 20% of positive samples when top 20% of cattle were chosen for a prediction task. In Figure 7, our DT model identified 54% of all positive samples, showing its performance between a random model and an ideal model. Based on this observation, the lift (or improvement) of DT over a random model is calculated as 54%/20% = 2.70, meaning that DT model performs 2.7 times better than a random model in terms of correctly identifying positive samples when top 20% of cattle were chosen for a prediction task. Similarly, when top 40% of cattle were chosen for a prediction task, the lift (or improvement) of DT over a random model is calculated as 83%/40% = 2.07.

We first noted that the supply chains in Jeollabuk-do province (left in Figure 8) are confined within its own geographical boundary. In addition, all supply chains in Jeollabuk-do province are associated with average quality of meat (i.e., between 2 and 3 denoted as yellow line). Based on these observations, we concluded that the meat quality produced in Jeollabuk-do province is an average level and it is most likely consumed within its province.

In contrast, supply chains in Jeollanam-do province (right in Figure 8) show all three levels of meat grades from high quality in red line, average quality in yellow lines, and low quality in red lines. In addition, logistical distances of these supply chains seem to be slightly longer than those in Jeollabuk-do province. However, supply chains in both Jeollabuk-do and Jeollanam-do are commonly defined as local in a sense that each of their geographical connections between farms and a slaughter house is limited within its own province.

6.2 Regional Supply Chains

We defined regional supply chains as supply chains between farms in neighboring provinces to a slaughter house, and we presented such supply chains from Gyeongsang Province in Figure 9.

Figure 9: Regional Supply Chains in Gyeongsang Province

We immediately noted that three supply chains shown in Figure 9 are different from those shown in Figure 8 in that these supply chains present connections between farms and slaughter houses across the boundary of Gyeongsang province, the southern eastern region of the Korean peninsula. For example, slaughter houses in the first two supply chains in Figure 9 are located in Gyeongsangnam-do (a southern part of Gyeongsang province) but have established supply chains with many farms in neighboring provinces such as Jeolla (Jeollanam-do & Jeollabuk-do), Chungcheong (the mid-west (Chungcheongnam-do) and mid-east region (Chungcheongbuk-do)),

Figure 10: National Supply Chains in Chungcheong Province

This supply chain clearly shows the characteristic of a national supply chain through its connection between associated slaughter house and farms beyond geographical boundary of neighboring provinces. This is somewhat expected in advance because Chungcheong province is the geographical hub of highway traffics in Korea due to its central location. As evidenced in Figure 10, the supply chain of the chosen slaughter house represents beef meat transportation from multiple farms in nationwide regions such as Jeolla, Gyeongsang, Chungcheong, and even Gangwon regions. However, it turns out that the quality of meat along this national supply chain is low quality.

Figure 11: National Supply Chains in Gyeonggi Province

As shown in Figure 11, all three supply chains in Gyeonggi Province found to be national supply chains, connecting slaughter houses to farms in nationwide. Again, this is mainly because Gyeonggi province is the closest to Seoul, the capital city of the Korea where the largest portion of beef meat is demanded and consumed in both wholesale and retail markets.

At the same time, we will focus on finding ways of enhancing the traceability systems so that both governmental authorities and businesses track and share information in beef quality certificate with consumers through various online sites.

References

1. Lim, D., Choi, B.H., Cho, Y.M., Chai, H.H., Jang, G.W., Gondro, C., Jeong. Y.H., & Lee, S.H. (2016). Analysis of extended haplotype in Korean cattle (Hanwoo) population. BMB Report, 49(9), 514-519. doi:10.5483/bmbrep.2016.49.9.074.

2. Lee, S.-H., Park, B.-H., Sharma, A., Dang, C.-G., Lee, S.-S., Choi, T.-J., Choy, Y.-H., Kim, H.-C., Jeon, K.-J., Kim, S.-D., Yeon, S.-H., Park, S.-B., & Kang, H.-S. (2014). Hanwoo cattle: Origin, domestication, breeding strategies and genomic selection. Journal of Animal Science and Technology, 56(2). doi:10.1186/2055-0391-56-2.

3. Kim, E.H., Cheong, H.S., Bae, J.S., Chun, J.-Y., Park, T.J., Lee, K.K., Yun, Y.M., & Shin, H.D. (2010). Identification of genetic polymorphisms in bovine mtDNA. Journal of Animal Science, 88(8), 2551-2555. doi:10.2527/jas.2009-2235.

4. Charlebois, S., Latif, N., Ilahi, I., Sarker, B., Music, J., & Vezeau, J. (2024). Digital traceability in agri-food supply chains: A comparative analysis of OECD member countries. Foods, 13(7), 1075. https://doi.org/10.3390/foods13071075.

5. Yang, C. (2022). Digital contact tracing in the pandemic cities: Problematizing the regime of traceability in South Korea. Big Data & Society, 9(1), https://doi.org/10.1177/20539517221089294.

6. Hong, E., Lee, S.Y., Jeong, J.Y., Park, J.M., Kim, B.H., Kwon, K., & Chun, H.S. (2017). Modern analytical methods for the detection of food fraud and adulteration by food category. Journal of the Science of Food and Agriculture, 97(12), 3877–3896. https://doi.org/10.1002/jsfa.8364.

7. Kamruzzaman, M., Makino, Y., & Oshita, S. (2015). Non-invasive analytical technology for the detection of contamination, adulteration, and authenticity of meat, poultry, and fish: A review. Analytica Chimica Acta, 853, 19-29. https://doi.org/10.1016/j.aca.2014.08.043.