Price Prediction Model For Ceramic Artworks Based On Machine Learning

Artwork price forecasting is a very important subject in the auction industry.However,the prices of artworks are susceptible to market changes and it is difficult to make accurate forecasts.As the art market continues to evolve,the instability and complexity of artwork prices is becoming more apparent.To address this problem,this paper explores a method for predicting the price of ceramic artworks based on machine learning algorithms.In this paper,the ceramic artwork auction data from the four major auction houses during the period 2017 to 2021 were pre-processed,including data cleaning.Then the data were subjected to exploratory analysis,descriptive and visual analysis.Finally,three methods of ANOVA,correlation coefficient method and model importance analysis were used for feature screening,and 11 important features,such as size and title recognition,were comprehensively screened out.For the study of ceramic artwork price prediction models,this paper constructs SVR,XGBoost and RFR models based on the training set after feature screening,and analyzes and compares the results.Subsequently,the 10-fold cross-validation method was used to optimize the model parameters.Finally,the model’s generalizability was verified by the independent testing method.The results of the independent test validation showed that the RFR model outperformed the SVR and XGBoost regression models with a MAPE value of 0.3159.The analysis of the prediction results showed that the RFR model was better at predicting medium to high priced ceramic artworks,but still had a large error in predicting the price of low-priced ceramics.Therefore,setting the median actual auction price of ceramic artworks at 230,000 yuan as the watershed of the independent test set,the RFR model still has the best prediction performance for the test set above 230,000 yuan,with a fit of 97.23%and a MAPE value of 0.2055,which has a high reference value;however,on the test set below 230,000 yuan,the RFR model has a poor prediction effect.It is inferred that the RFR model can be extended and applied to the price prediction and analysis of medium to high priced ceramic artworks.Finally,according to the findings of this paper,corresponding recommendations are made for low-priced and medium to high priced ceramic artworks.Based on the shortcomings of data collection scope,data reliability and model applicability,the future research direction is pointed out.