search
Contact Us
HOME

  • Crossref logo
  • Crossref Similarity Check logo
Journal Search Engine

to

Search Advanced Search Adode Reader(link)
Download PDF Export Citaion korean bibliography PMC previewer
ISSN : 2287-5824(Print)
ISSN : 2287-5832(Online)
Journal of The Korean Society of Grassland and Forage Science Vol.46 No.3 pp.241-248
DOI : https://doi.org/10.5333/KGFS.2026.46.3.241

Selection of Major Climate Variables Affecting Alfalfa Dry Matter Yield and Its Prediction Using mRMR and Decision Tree Model

Jiyung Kim1, Kyung Il Sung2, Jae Seong Choi3, Byong Wan Kim3*
1Forages Production Systems Division, National Institute of Animal Science, Cheonan 31000, Republic of Korea
2Department of Animal Industry Convergence, Kangwon National University, Chuncheon 24341, Republic of Korea
3Department of Animal Science, Kangwon National University, Chuncheon 24341, Republic of Korea
*Corresponding author: Byong Wan Kim, Department of Animal Life Science, Kangwon National University, Chuncheon 24341, Republic of Korea. Tel: +82-33-250-8625, Email: bwkim@kangwon.ac.kr
September 22, 2026 September 28, 2026 September 29, 2026

Abstract


This study aimed to identify the major climatic variables affecting dry matter yield (DMY) of alfalfa (Medicago sativa L.) and to predict DMY using a decision tree model. A total of 235 observations of alfalfa DMY collected from field experiments conducted in South Korea were combined with climate data obtained from the Korea Meteorological Administration. Twenty-six climate variables related to temperature, growing days, growing degree days, precipitation, duration of sunshine, and summer depression were considered as candidate variables. The minimum Redundancy Maximum Relevance (mRMR) algorithm was applied to select climate variables with high relevance to DMY and low redundancy among variables. Six variables, including growing degree days from 0 to 25°C (GDD 0–25), growing days from 5 to 25°C (GD 5–25) and from 0 to 25°C (GD 0–25), accumulated temperature of summer depression (ASD), duration of sunshine from 5 to 25°C (DS 5–25), and days with precipitation ≥5 mm (DP 5) were selected. The decision tree model showed an R2 of 0.8103 and an RMSE of 2,804.0 kg ha−1. GD 0–25 was selected as the root-node variable, with 266 days as the primary split point. Eight terminal nodes were generated, with predicted DMY ranging from 3,620 to 24,647 kg ha−1. The highest predicted DMY was observed under conditions of GD 0–25 ≥266 days and DP 5 ≥60 days, whereas the most frequently observed prediction was 18,364 kg ha−1, accounting for 39% of the observations. These results indicate that alfalfa DMY is associated with conditional combinations of growing days, the number of days with precipitation, accumulated temperature, and sunshine duration. The mRMR-based decision tree approach may provide a useful framework for identifying major climate variables and interpreting threshold-dependent responses of alfalfa productivity.


Alfalfa , Dry matter yield , Climate variables , Decision tree model , mRMR algorithm

초록

    Ⅰ. INTRODUCTION

    Alfalfa (Medicago sativa L.) is a perennial forage legume characterized by high dry matter yield (DMY) and crude protein concentration (Undersander et al., 2011). Its productivity is influenced by cultivar and crop management, as well as environmental conditions, including temperature, precipitation, sunshine duration, and soil physical and chemical properties (Baral et al., 2022;Bourgeois et al., 1990;Kim et al., 2021). Because alfalfa undergoes repeated regrowth and harvest after overwintering, annual DMY may reflect the combined effects of accumulated climatic conditions and seasonal stresses rather than weather conditions at a single growth stage. In particular, water stress is a major constraint on photosynthesis and biomass production (Baral et al., 2022;Diatta et al., 2021;Undersander, 2021), while high temperature can also affect alfalfa physiological functioning and productivity (Undersander, 2021).

    Previous studies have quantified relationships between environmental conditions and alfalfa productivity in South Korea. Kim et al. (2021) developed yield prediction models using domestic field-trial data and identified growing degree days (GDD) from 0 to 25℃ of mean air temperature and growing days (GD) above 0℃ of mean air temperature as major climatic variables. The climate-only multiple regression model had an R2 of 0.56, and a general linear model incorporating soil clay content provided greater explanatory power. The reported contributions of accumulated temperature and growing days were 21% and 11%, respectively, indicating the importance of thermal conditions and growing-period length. Kim et al. (2023) evaluated 197 domestic alfalfa observations and identified three major environmental factors; their regression model had an R2 of 0.602. Temperature accumulation, conditions above 25℃, sunshine duration within 0-25℃, and precipitation-related variables were important components.

    These studies provide quantitative descriptions of environmental associations with alfalfa DMY. However, the particular linear models used primarily describe average relationships and may not explicitly reveal thresholds or condition-dependent responses unless nonlinear terms or interactions are specified. This limitation should not be generalized to all statistical models. Moreover, variables generated from climate records can be strongly correlated. Growing days, thermal accumulation, and sunshine duration can each be calculated using several temperature thresholds, resulting in variables that contain overlapping information. Selecting variables that are relevant to DMY while limiting redundancy may therefore improve the interpretability of the analysis.

    The minimum redundancy maximum relevance (mRMR) is a feature-selection method that seeks high relevance to the response while limiting redundancy among variables (Peng et al., 2005). It can be useful when numerous correlated variables are derived from the same meteorological records using different thresholds. Nevertheless, mRMR does not guarantee statistical independence among the selected variables.

    A decision tree is a nonparametric supervised learning method that recursively partitions observations according to variable thresholds. For a continuous response such as DMY, the resulting model is a decision tree. At each split, the model selects a variable and threshold that reduce within-node variation, and each terminal node provides a prediction for observations satisfying the corresponding conditions (Breiman et al., 1984). Decision trees do not require a prespecified linear relationship and provide an interpretable representation of conditional patterns through a tree structure and if-then rules.

    Accordingly, this study compiled alfalfa DMY observations and corresponding climate data from field experiments conducted in South Korea. The objectives were to select relevant climatic variables with limited redundancy using mRMR, construct a decision tree to predict DMY from their combinations, and evaluate model performance. The resulting split points were interpreted as empirical thresholds within the available data.

    Ⅱ. MATERIALS AND METHODS

    1. Data collection and processing

    A total of 318 alfalfa data were compiled from adaptability test of imported varieties of grasses and forage crops operated by National Agricultural Cooperative Federation(NACF) (69 data), research reports on livestock experiments operated by Korean National Livestock Research Institute (192 data), a research paper in Journal of the Korean Society of Grassland and Forage Science (KSGFS) (30 data), research paper in Journal of Animal Science and Technology (JAST) (1 data), and academic theses (26 data). Climate records were obtained from the Korea Meteorological Administration Open MET Data Portal (KMA, 2026) and matched to the experimental location and cultivation year. When a meteorological station was located within the experimental area, its records were used; otherwise, records from the nearest station by straight-line distance were selected. Daily temperature, precipitation, and sunshine duration were used to derive variables representing thermal conditions, precipitation, sunshine, and seasonal temperature stress.

    Data processing comprised four stages. First, alfalfa and climate records were entered into spreadsheets. Second, the records were screened for duplicates, transcription errors, and calculation errors. Third, the crop and climate records were merged and climatic variables were generated. Finally, ineligible or incomplete records were removed. The stated exclusions comprised 10 duplicate data, 33 spring-sown data, nine data from paddy fields, 28 data without harvest-frequency information, and three data without yield data. The final dataset contained 235 observations: 27 from Namwon, 26 from Dangjin, 118 from Suwon, 15 from Icheon, 6 from Cheonan, and 43 from Pyeongchang.

    2. Derivation of climatic variables

    Twenty-six climatic variables were generated, including GD, GDD, sunshine duration (DS), precipitation days (DP), accumulated precipitation (AP), days associated with summer depression (DSD), accumulated temperature associated with summer depression (ASD), mean January temperature (MTJ), and minimum January temperature (LTJ) (Table 1).

    GD 0-25 and GD 5-25 represented the numbers of days with daily mean temperatures within 0-25℃ and 5-25℃, respectively. Nine GDD variables were constructed using lower temperature thresholds of 0, 5, or 10℃, with no upper limit or an upper limit of 25 or 30℃. Seven DS variables were derived using the temperature conditions denoted by 0, 0-25, 0-30, 5, 5-25, 5-30, and 10. Sunshine duration was summed over days satisfying the corresponding temperature condition.

    Precipitation variables were derived for days with daily mean temperatures of at least 0℃. DP 0 and DP 5 described precipitation-day counts under the respective rainfall thresholds, while AP 0 and AP 5 described the corresponding accumulated rainfall. DP 5 and AP 5 used a daily rainfall threshold of at least 5 mm.

    ASD was defined in the source as the sum of daily mean temperatures on days with daily mean temperature at least 25℃, and DSD as the number of such days. MTJ and LTJ represented mean and minimum January temperatures, respectively. These temperature variables were used as climatic indicators of summer heat exposure and winter conditions, rather than direct measurements of physiological depression or winter hardiness.

    3. Variable selection and model evaluation

    The mRMR algorithm selected climatic variables by balancing relevance to alfalfa DMY against redundancy among variables. For a variable set S and response Y, average relevance and average redundancy were denoted by D ( S , Y ) and R ( S ) , respectively. Variables were selected sequentially using the difference criterion, D ( S , Y ) − R ( S ) (Peng et al., 2005). Because mRMR does not provide a model-based variable importance measure comparable to that of a decision tree, the consistency of variable selection was summarized using selection frequency (%). The six climatic variables with the highest selection frequencies were retained as candidate explanatory variables for the decision tree.

    At each node, the decision tree selected the climatic variable and threshold that maximized the reduction in residual sum of squares. The maximum tree depth was limited to four to control complexity and maintain interpretability. The predicted DMY in each terminal node was the mean DMY of the observations assigned to that node.

    The dataset was randomly divided into a training set containing 80% of the observations (n = 191) and a held-out test set containing 20% (n = 44). The training set was used for variable selection and model fitting. The test set was reserved for evaluating the final fitted model and was not used for variable selection or model development.

    Five-fold cross-validation was conducted within the training set. The observations were divided into five nonoverlapping subsets. In each iteration, four folds were used for fitting and the remaining fold for validation. This procedure was repeated until each fold had served once as the validation set. Performance was assessed using R2 and root mean squared error (RMSE). The final decision tree was fitted using the full training set and evaluated on the held-out test set.

    Variable importance was calculated to compare the contributions of the variables to the fitted decision tree and expressed as a percentage of the sum of the importance values.

    4. Statistical software

    Data preprocessing, climatic variable selection, and decision tree analysis were performed using R version 4.4.1 (R Core Team, 2024) in RStudio. The decision tree was fitted using the rpart package and visualized using rpart.plot. Model performance was evaluated using R2 and RMSE.

    Ⅲ. RESULTS AND DISCUSSION

    1. Descriptive statistics

    Descriptive statistics for alfalfa DMY and the climatic variables are presented in Table 2. Mean and median of DMY were 14,001.7 and 14,617.0 kg ha−1, respectively. Mean of GD 0-25 and GD 5-25 were 276.5 and 221.3 days, a difference of approximately 55 days. Mean of GDD 0, GDD 5, and GDD 10 were 4,300, 3,000, and 1,900℃, respectively. Mean of DP 0 and DP 5 were 102 and 47 days, while mean of AP 0 and AP 5 were 1,318 and 1,229 mm, a difference of approximately 90 mm. Mean of DS 0, DS 5, and DS 10 were 1,700, 1,400, and 1,200 hr, respectively. Mean of ASD and DSD were around 693℃ and 25 days, while mean of MTJ and LTJ were -3.12 and -7.99℃.

    2. Selection frequency of climatic variables using the mRMR algorithm

    The mRMR analysis showed that GDD 0–25 had the highest selection frequency, followed by GD 5–25 and GD 0–25, ASD, DS 5–25, and DP 5 (Fig. 1). In contrast, the relative variable importance derived from the decision tree was highest for GD 0–25, followed by ASD, GD 5–25, GDD 0–25, DP 5, and DS 5–25 (Fig. 2).

    Differences between mRMR selection frequency and decision tree variable importance are expected because these metrics represent different aspects of variable relevance. The mRMR selection frequency reflects how consistently a variable is retained during the feature-selection procedure while balancing relevance to DMY against redundancy with other variables. In contrast, decision tree importance reflects the contribution of a variable to reducing variation in DMY within the fitted tree. Therefore, a variable with a high mRMR selection frequency does not necessarily have high variable importance in the decision tree.

    GD 0–25, GD 5–25, and GDD 0–25 were derived from the same temperature records and may therefore retain partially overlapping information even after mRMR selection. Such redundancy can influence how frequently individual variables are selected by mRMR, because correlated variables may provide similar information about DMY. In the decision tree, however, their relative importance depends on how effectively each variable partitions DMY within particular subsets of the data.

    3. Climatic variables affecting alfalfa dry matter yield and their effects as assessed using a decision tree model

    The R2 and RMSE for the independent test set were 0.8103 and 2,804.0 kg ha−1, respectively. GD 0-25 was selected for the root split, dividing the observations at 266 days (Fig. 3). Subsequent splits involved ASD, DP 5, GDD 0-25, GD 0-25, and DS 5-25. Eight terminal nodes had predicted DMY values ranging from 3,620 to 24,647 kg ha−1 and contained approximately 4-39% of the observations used to fit the displayed tree.

    The selection of GD 0-25 at the root indicates that, among the variables available to the fitted tree, this variable provided the largest improvement in the first partition. Previous research identified growing days and accumulated temperature as important correlates of alfalfa productivity (Kim et al., 2021, 2023). In the present tree, observations with GD 0-25 below 266 days generally received lower DMY predictions, whereas several higher-yield terminal nodes occurred when GD 0-25 was at least 266 days. Longer periods with temperatures compatible with growth may support repeated regrowth and harvest. However, the 266-day split identified by the decision tree is a data-derived partition of the observed data, not an experimentally established physiological optimum. It should therefore be interpreted as an exploratory, dataset-specific split within the observed range of day of year, rather than as a universal threshold (Venkatasubramaniam et al., 2017;Qian, 2014).

    When GD 0-25 was below 266 days, ASD provided the next split. Predicted DMY was 14,574 kg ha−1 when ASD was at least 938℃. When ASD was below 938℃, a further split at 13℃ separated predicted yields of 3,620 and 8,578 kg ha−1. Summer temperature conditions therefore distinguished additional variation within this subset. This pattern should not be interpreted as evidence that increasing heat stress improves alfalfa productivity (Baron and Bélanger, 2007;Matthews et al., 2018;Teuber et al., 1988). ASD could be associated with other location- or year-specific conditions, and its relationship with yield is conditional on the preceding GD split. Kim et al. (2023) also identified variables related to summer depression variable as major climatic factors associated with alfalfa productivity, indicating that the relationship between temperature and DMY should be interpreted in conjunction with other climatic conditions. Therefore, the effect of ASD observed in the present study is more appropriately interpreted as a conditional relationship occurring under the higher-level condition of GD 0–25 <266 days, rather than as a direct positive effect of high temperature.

    When GD 0-25 was at least 266 days, DP 5 provided the next split. The highest predicted DMY, 24,647 kg ha−1, occurred when DP 5 was at least 60 days. This terminal node contained 12 observations, approximately 6% of the fitting sample displayed in Fig. 3. The frequency of rainfall events of at least 5 mm may convey information different from accumulated rainfall. Alfalfa develops a relatively deep root system and is generally tolerant to water deficit; however, prolonged water limitation can reduce photosynthesis and biomass production and is a major environmental constraint on alfalfa productivity (Diatta et al., 2021;Undersander, 2021). Thus, more frequent precipitation events above a certain threshold may have alleviated prolonged water limitation and provided adequate water for regrowth following repeated harvests, thereby favoring annual DMY production. Nevertheless, the terminal node with DP 5 ≥60 days represented only 6% of the observations; therefore, this threshold should not be generalized as an optimal condition for achieving the highest DMY. Moreover, because soil drainage and actual soil moisture were not included in the model, a greater number of precipitation days should not be interpreted as necessarily leading to higher DMY.

    Within observations satisfying GD 0-25 ≥266 days and DP 5 <60 days, GDD 0-25 divided the data at 4,765℃. This indicates that accumulated temperature supplied additional information beyond the duration of the temperature-defined growing period. GDD represents the accumulated temperature above a defined base temperature and is therefore closely related to crop growth and development period. Kim et al. (2021) also identified GDD 0–25 as a major climatic variable affecting alfalfa DMY in South Korea, with a relatively higher standardized coefficient than GD in the regression model. Therefore, the selection of GDD 0–25 at an internal node of the present decision tree indicates that not only the length of the growing period but also the accumulated temperature during that period is important for explaining variation in alfalfa DMY.

    When GDD 0-25 was at least 4,765℃, a further GD 0-25 split occurred at 281 days. Predicted DMY was 10,146 kg ha−1 when GD 0-25 was at least 281 days and 15,122 kg ha−1 when it was at least 266 but below 281 days. Thus, the fitted tree did not represent yield as increasing monotonically with GD. Under conditions of high GDD, a very long growing period may coincide with prolonged exposure to high temperatures or occur in combination with other environmental conditions that vary among locations and years. Therefore, the lower predicted DMY observed under GD 0–25 ≥281 days should not be interpreted simply as a direct negative effect of a longer growing period on productivity. Rather, it is more appropriately interpreted as a conditional pattern observed under the preceding condition of GDD 0–25 ≥4,765℃.

    When GDD 0-25 was below 4,765℃, DS 5-25 separated predicted DMY at 1,293 hr. Predicted DMY was 12,766 kg ha−1 below that sunshine threshold and 18,364 kg ha−1 at or above it. The latter node contained 74 observations, approximately 39% of the displayed fitting sample, and was the largest terminal node. Sunshine duration is associated with opportunities for photosynthesis, although it is not equivalent to measured solar radiation. The result is consistent with a conditional association between sunshine and DMY. Kim et al. (2023) also identified a sunshine-related variable, although their cited variable was DS 0-25 rather than DS 5-25. In the present study, relatively high DMY was observed when DS 5–25 was high under conditions of a sufficiently long growing period, a limited number of precipitation days, and GDD 0–25 <4,765℃, suggesting that sunshine duration served as an additional variable differentiating alfalfa productivity. Notably, the terminal node with a predicted DMY of 18,364 kg ha−1 accounted for 39% of the observations. This terminal node contained a larger proportion of the observations in the dataset.

    Ⅳ. CONCLUSION

    The mRMR and decision tree results showed that alfalfa DMY varied according to combinations of growing days, precipitation frequency, accumulated temperature, summer depression conditions, and sunshine duration. Although GD 0–25 was selected as the primary splitting variable, different DMY values were predicted depending on the levels of DP 5 and GDD 0–25 even under similar GD conditions, while GD 0–25 and DS 5–25 further differentiated DMY at lower nodes. Therefore, the split points identified in this study should be interpreted as empirical thresholds that effectively partitioned variation in DMY within the compiled domestic alfalfa production data, rather than as physiological optima for individual climate variables. In addition, because the pathways identified by the decision tree represent conditional associations rather than causal relationships, further studies should evaluate the relationships between climatic factors and alfalfa productivity using datasets that also include soil properties, cultivar, harvest frequency, and other management factors.

    Figure

    KGFS-46-3-241_F1.jpg

    Selection frequency of climatic variables using the mRMR algorithm for alfalfa dry matter yield prediction.

    Note: GDD 0-25: Growing Degree Days 0-25; GD 5-25: Growing Days 5-25; GD 0-25: Growing Days 0-25; ASD: Accumulated temperature of Summer Depression; DS 5-25: Duration of Sunshine 5-25; DP 5: The number of Days with Precipitation 5.

    KGFS-46-3-241_F2.jpg

    Relative importance of climatic variables in the decision-tree model for alfalfa dry matter yield.

    Note: GD 0-25: Growing Days 0-25; ASD: Accumulated temperature of Summer Depression; GD 5-25: Growing Days 5-25; GDD 0-25: Growing Degree Days 0-25; DP 5: The number of Days with Precipitation 5; DS 5-25: Duration of Sunshine 5-25.

    KGFS-46-3-241_F3.jpg

    Selection of climate variables and predicted alfalfa dry matter yield (kg ha−1) as a decision tree model.

    Note: GD 0-25: Growing Days 0-25; ASD: Accumulated temperature of Summer Depression; DP 5: The number of Days with Precipitation 5; GDD 0-25: Growing Degree Days 0-25; GD 5-25: Growing Days 5-25; DS 5-25: Duration of Sunshine 5-25.

    Table

    Definitions and units of the climatic variables considered for alfalfa dry matter yield prediction
    Descriptive statistics for alfalfa dry matter yield and climatic variables in the compiled field-trial dataset from 1982 to 2022
    Note: DMY: Dry matter yield; GD 0-25 and 5-25: Growing Days 0-25, 5-25; GDD 0, 0-25, 0-30, 5, 5-25, 5-30, 10, 10-25, and 10-30: Growing Degree Days 0, 0-25, 0-30, 5, 5-25, 5-30, 10, 10-25, and 10-30; DP 0 and 5: The number of Days with Precipitation 0, 5; AP 0 and 5: Amount of Precipitation 0, 5; DS 0, 0-25, 0-30, 5, 5-25, 5-30, and 10: Duration of Sunshine 0, 0-25, 0-30, 5, 5-25, 5-30, and 10; ASD: Accumulated temperature of Summer Depression; DSD: Days of Summer Depression; MTJ: Mean Temperature on January; and LTJ: Lowest Temperature on January.

    Reference

    1. Baral, R., Lollato, R.P., Bhandari, K. and Min, D. 2022. Yield gap analysis of rainfed alfalfa in the United States. Frontiers in Plant Science. 13:931403. (doi:)
    2. Baron, V.S. and Bélanger, G. 2007. Climate and forage adaptation. In: R.F. Barnes, C.J. Nelson, K.J. Moore and M. Collins (Eds.), Forages: The science of grassland agriculture. Blackwell Publishing. Ames, IA, USA. pp. 83-104.
    3. Bourgeois, G., Savoie, P. and Girard, J.M. 1990. Evaluation of an alfalfa growth simulation model under Québec conditions. Agricultural Systems. 32:1-12. (doi:)
    4. Breiman, L., Friedman, J.H., Olshen, R.A. and Stone, C.J. 1984. Classification and regression trees. Wadsworth International Group. Belmont, CA, USA.
    5. Diatta, A.A., Min, D. and Jagadish, S.V.K. 2021. Drought stress responses in non-transgenic and transgenic alfalfa—Current status and future research directions. Advances in Agronomy. 170:35-100. (doi:)
    6. Kim, J.Y., Kim, M.J., Jo, H.W., Lee, B.H., Jo, M.H., Kim, B.W. and Sung, K.I. 2021. Assessment of contribution of climate and soil factors on alfalfa yield by yield prediction model. Journal of the Korean Society of Grassland and Forage Science. 41(1):47-55. (doi:)
    7. Kim, J.Y., Han, K.J., Sung, K.I., Kim, B.W. and Kim, M. 2023. Assessment of growing condition variables on alfalfa productivity. Journal of Animal Science and Technology. 65(5):939-950. (doi:)
    8. KMA. 2026. KMA weather data service, open MET data portal. Korea Meteorological Administration. Seoul, Korea. https://data.kma.go.kr/
    9. Matthews, C., Arshad, M. and Hannoufa, A. 2018. Alfalfa response to heat stress is modulated by microRNA156. Physiologia Plantarum. 165(4):830-842. (doi:)
    10. Peng, H., Long, F. and Ding, C. 2005. Feature selection based on mutual information: Criteria of max-dependency, max-relevance, and min-redundancy. IEEE Transactions on Pattern Analysis and Machine Intelligence. 27(8):1226-1238. (doi:)
    11. Qian, S.S. 2014. Ecological threshold and environmental management: A note on statistical methods for detecting thresholds. Ecological Indicators. 38:192-197. (doi:)
    12. R Core Team. 2024. R: A language and environment for statistical computing. R Foundation for Statistical Computing. Vienna, Austria. https://www.R-project.org/
    13. Teuber, L., Jernstedt, J. and Foord, K. 1988. Alfalfa growth and development. Proceedings of the 18th California Alfalfa Symposium. Modesto, CA, USA.
    14. Undersander, D. 2021. Economic importance, practical limitations to production, management, and breeding targets of alfalfa. In: L.X. Yu and C. Kole (Eds.), The alfalfa genome. Springer. Cham, Switzerland. pp. 1-11. (doi:)
    15. Undersander, D., Cosgrove, D., Cullen, E., Grau, C., Rice, M.E., Renz, M., Sheaffer, C., Shewmaker, G. and Sulc, M. 2011. Alfalfa management guide. American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America. Madison, WI, USA.
    16. Venkatasubramaniam, A., Wolfson, J., Mitchell, N., Barnes, T., JaKa, M. and French, S. 2017. Decision trees in epidemiological research. Emerging Themes in Epidemiology. 14:11. (doi:)
    Copyright © The Korean Society of Grassland and Forage Science. All rights reserved.