Lasso 기계학습을 이용한 주요 어종의 소비자가격 예측모형에 관한 연구

A Study on the Forecasting Models for Consumer Prices of Major Fishery Species Using Lasso Machine Learning

This study constructed the prediction models of frozen and fresh consumer prices for mackerel and hairtail based on Lasso and its variants as a machine learning technique in artificial intelligence and tested the statistical differences in forecasting power among the models. In the case of mackerel, Lasso model was selected as the best model for the frozen consumer price on the basis of the MSE, and Sqrt-Lasso model was the best on both MAE and MAPE. Adaptive-Lasso model was the best model for the fresh consumer price on the basis of all of the three criteria, MSE, MAE, and MAPE. In the case of hairtail, on the basis of MSE, Lasso model was selected as the best model, just like the case of mackerel, and Adaptive-Lasso model was the best on the basis of both MAE and MAPE. Lasso model was the best model for the fresh consumer price on the basis of all of the three criteria. However, there was no statistical significance found in the differences in forecasting power of the models among all of the comparison tests using a paired t-test between the best model and the competitive model. And hence, as long as the power of prediction of consumer prices is concerned, all of the three Lasso models utilized in this study are turned out to be the same.