작물 생산량 예측을 위한 머신러닝 기법 활용 연구

국문 초록 (Abstract)

기후변화로 인해 농작물 생산에 대한 관심이 증대되고 있고, ICT 기술을 접목한 스마트팜의 작물 생산량을 최적화하기 위한 노력이 증대되고 있다. 빅데이터를 활용한 작물과 환경 상태에 따른 생산 정도의 분석이 필요하고 나아가 생산량 예측을 위한 모델에 대한 연구가 필요하다. 3가지의 머신러닝 알고리즘 (Ridge Regression, Random Forest, XGBoost)을 후보 알고리즘으로 선정하여 작물 생산량 예측의 적합도를 평가 분석하였다. 실제값과 예측값의 오차를 산출한 MAE(Mean Absolute Error)/RMSE(Root Mean Square Error)값을 모델 평가 지표로 사용하여 알고리즘별 최적의 파라미터를 선정하였다. Ridge Regression의 파라미터 λ는 2.512이고, Random Forest의 파라미터는 분할 8, 트리 100이고, XGBoost의 파라미터는 감마 0, 깊이 10 이였다. 선정된 최적의 파라미터가 결정된 알고리즘들을 MAE/RMSE로 평가한 결과 XGBoost가 MAE 0.233, RMSE 0.817로 최소값을 나타내 최적 모델로 선정되었고, 변수 중요도를 확인해 본 결과, 재식밀도, 생장길이, 잎의 수 요인이 출하량 예측에 가장 중요한 요인으로 도출되었다. XGBoost 모델의 예측력을 R-Square 값을 통해 평가하였고, 각 작기별 총 출하량 예측 모델에 대하여 약 77%의 설명력을 보였다.

다국어 초록 (Multilingual Abstract)

Climate change has increased interest in crop production and efforts to optimize crop production of smart farms incorporating ICT technology. Therefore, it is necessary to analyze the production level according to the crops and environmental conditions using big data. Hence, a model for predicting production is required. Three machine learning algorithms (Ridge Regression, Random Forest, and XGBoost) were selected as candidate algorithms to evaluate and analyze the fit of predicting crop production. The MAE (Mean Absolute Error) / RMSE (Root Mean Square Error) values, which calculated the error between the actual and predicted values, were used as model evaluation indices. The parameters with the smallest MAE/RMSE values were selected for the optimal model of each algorithm. Parameter λ of Ridge Regression was 2.512, the parameters of the Random Forest were division 8 and tree 100, and the parameters of XGBoost were gamma 0 and depth 10. An evaluation of the algorithms by MAE/RMSE revealed XGBoost to be the optimal model with minimum values of MAE = 0.233 and RMSE = 0.817. An examination of the importance of the variables revealed the planting density, growth length, and number of leaves to be important factors in forecasting shipments. The predictive power of the XGBoost model was evaluated using the R-Square value, which had an explanatory power of approximately 77% for the total shipment forecast model for each production period.

목차 (Table of Contents)

• 요약
• Abstract
• 1. 서론
• 2. 본론
• 3. 결론
• References

참고문헌 (Reference)

• 1 Vishal Morde, "XGBoost Algorithm"
• 2 "United Nations Framework Convention on Climate Change"
• 3 Inha University, "Trends and Prospects of the Climate Change Convention"
• 4 Rokach, L., "Top-down induction of decision trees classifiers-a survey" 2005
• 5 Osisanwo F.Y, "Supervised Machine Learning Algorithms" Classification and Comparison
• 6 "Smart Farming Technology"
• 7 Ministry of Agriculture, Food and Rural Affairs, Republic of Korea, "Smart Farm Dispersion Method"
• 8 Saptashwa Bhattacharyya, "Ridge and Lasso Regression: L1 and L2 Regularization"
• 9 Antonio Criminisi, "Regression Forests for Efficient Anatomy Detection and Localization in CT Studies"
• 10 Jamie Shotton, "Real-time human pose recognition in parts from single depth images"
• 11 Dev Cristoval, "Random Forest Algorithm"
• 12 "Predictive algorithms and AI"
• 13 TO Ayodele, "New advances in machine learning" 2010
• 14 "Nature Climate Change"
• 15 Baek Kyun Shin, "Machine Learning-Decision Tree"
• 16 Ayon Dey, "Machine Learning Algorithms: A Review"
• 17 Mohssen Mohammed, "Machine Learning Algorithms and Applications"
• 18 "Korea Policy Briefing"
• 19 "IPCC Special Report on the Impacts of Global Warming of 1.5 °C"
• 20 "IPCC — Intergovernmental Panel on Climate Change"
• 21 Jenna Burrell, "How the machine ‘thinks’: Understanding opacity in machine learning algorithms"
• 22 Kursa M, "Feature Selection with the Boruta Package" 36 (36): 2010
• 23 Darko Zikic, "Decision Forests for Tissue-Specific Segmentation of High-Grade Gliomas in Multi-channel MR"
• 24 "Datacamp tutorial"
• 25 "Climate change –"
• 26 Meteorological Agency, "Climate Information Portal"
• 27 NASA, "Climate Change and Global Warming"
• 28 Korea Environment Corporation, "Climate Change Public Relations Portal"
• 29 Papagelis A., "Breeding Decision Trees Using Evolutionary Techniques" 393-400, 2001
• 30 Korea University Business School, "Big Data Use Case and Implications"
• 31 Deng,H, "Bias of importance measures for multi-valued attributes and solutions" 293-, 2011
• 32 Chipman, Hugh A, "Bayesian CART model search" 93 (93): 1998
• 33 Murthy S, "Automatic construction of decision trees from data"
• 34 "Analytics Vidhya"
• 35 K Anderson, "A new paradigm for climate change" 2012
• 36 Susmita Ray, "A Quick Review of Machine Learning Algorithms"