Analysis of visitor data prediction of National Science Museum Thailand (NSM)

At present, many museums have begun to use data analysis for assisting management. For example, data prediction has been used to predict the number of visitor and help museum curators planning how to improve their service. In this thesis, we investigated and analyzed the factors that affect the number of visitors by using data correlation to select the factors. We proposed museum visitor number prediction method based on the Random Forest (RF) model and the XGBoost model using historical data from 2016 to 2020 from three museums under the supervision of the National Science Museum Thailand. Since all selected museums locate in the same area, to improve the prediction performance, our purposed method combines the data from the three museums. More specifically, the result of one museum is used to help predict the visitor number of the other museum. Originally, the RF and XGBoost models have average accuracy at 56.99% and 58.08%, respectively. We had proposed four methods in this thesis all of which have accuracy more than the original. Regarding to our proposed method, the highest average accuracy was 85.14 and 83.13% from proposed method D. We also observed that the XGBoost model consumed less computational time. In addition, the RF model performed better than the XGBoost model with 97.85% accuracy after reducing the visitor number classes from five to three.