대학 구내식당 식사 인원 예측을 위한 머신러닝 기반 모델 및 웹 앱 개발
Development of Machine Learning-Based Predictive Models and a Web App for University Cafeteria Attendance
2
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
정확한 식사 인원 예측은 잔반 감소와 원가 절감 등 식당 운영에 긍정적인 영향을 미친다. 과거에는 사람의 경험에 의존하여 예측했으나, 현대에는 예측 모델을 개발하여 사용하고 있다. 이러한예측 방법은 과거의 방식보다 적은 시간과 자원을 소모하는 등의 이점이 있다. 그러나 여러 어려움으로 인해 실제로 이를 시도하는 곳은 많지 않다. 이전 연구에서의 모델을 고도화하여 2018년, 2019년, 2023년 학기의 교내 식당 데이터와 기상청의 날씨 데이터를 사용하여 EDA 분석을 통해 변수를 선정하였다. 머신러닝 알고리즘을 사용하여 조식, 중식, 석식의 식사 인원을 예측하는 모델을개발하였고, 전체 데이터의 90%로 학습하고 나머지 10%를 검증에 사용하였다. 가장 성능이 우수한모델을 최종 모델로 선정하여 예측 프로그램을 개발하였다. 최종 모델은 조식 예측에XGBoost(MAE 비율 12.97%), 중식과 석식 예측에 RandomForest(MAE 비율 각각 4.8%, 4.93%)를 사용하였다. 중식과 석식 예측 모델이 좋은 성능을 보였다. 향후 지속적인 데이터 수집, 새로운 파생변수 추가, 하이퍼파라미터 튜닝, UI 개선을 통해 성능과 편의성을 더욱 향상시킬 예정이다.
Abstract
Accurate meal attendance prediction positively impacts cafeteria operations by reducing food waste and lowering costs. In the past, predictions were based on human experience, but modern methods involve developing predictive models. These models offer advantages such as reduced time and resource consumption compared to traditional methods. However, practical implementation remains limited due to various challenges. This paper utilizes cafeteria data from 2018, 2019, and 2023 semesters, along with weather data from the Korea Meteorological Administration, to select variables through EDA analysis. Using these variables and machine learning algorithms, models were developed to predict breakfast, lunch, and dinner attendance. The best-performing model was selected as the final model to develop a prediction program. 80% of the data was used for training and 20% for validation. The final models selected were XGBoost for breakfast (MAE ratio 12.97%) and RandomForest for lunch and dinner (MAE ratios 4.8% and 4.93%, respectively). The lunch and dinner prediction models demonstrated good performance. Future work will involve continuous data collection, adding new derived variables, hyperparameter tuning, and UI improvements to enhance model performance and convenience.
참고문헌 (18)
- Agnieszka Lasek, "Restaurant Sales and Customer Demand Forecasting : Literature Survey and Categorization of Methods", Smart City 360°, pp.479~491 (2016)
- J. Jeon, "Forecasting Models for Number of Persons for Meals in Large Food Service Place : Based on Actual Data of the S City Hall Restaurant for Employees", PhD, pp.2019
- L. Chung, "The development of a forecasting model as a management strategy in university foodservices", Master, pp.2001
- L. Chung, "Investigation on the Performance of the Forecasting Model in University Foodservice", The Korean Journal of Nutrition, 36(9), pp.966~973 (2003)
- H. J. Lim, "Comparative Assessment of Forecasting Models Applicable for Business and Industry Foodservice Operations", Master, pp.2008
- U. Kwon, "Analysis of priority factors in demand forecasting by cafeteria service providers : Focused on large-scale industrial cafeteria service providers", Master, pp.2020
- K. O. Cullen, "Menu Item Forecasting Systems in Hospital Foodservice", Journal of the American Dietetic Association, 73(6), pp.640~646 (1978)
- K. Jung, "The Development of Prediction Models for the Number of People for Meal at University Cafeteria", Proceedings of the Korean Society of Computer Information Conference, 31(2), pp.535~536 (2023)
- "KMA-Open MET Data Portal"[URL]
- Y. J. Jang, "The Change of The Average Discomfort Index from June to September during The Past 10 Years", Journal of Climate Change Research, 3(2), pp.89~100 (2012)
- E. C. Thom, "The Discomfort Index", Weatherwise, 12(2), pp.57~61 (1959)[DOI]
- P. Schneider, "Anomaly Detection and Complex Event Processing over IoT Data Streams With Application to eHealth and Patient Data Monitoring", pp.59~60 (2022), Academic Press
- Hodson, T. O., "Root-mean-square error (RMSE) or mean absolute error (MAE): when to use them or not", Geosci. Model Dev, 15, pp.5481~5487 (2022)[DOI]
- Chicco D, "The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluation", PeerJ Comput Sci, 7, pp.e623~ (2021)[DOI]
- "MLflow"[URL]
- "FastAPI"[URL]
- "Kafka"[URL]
- "Grafana"[URL]