A Comparative Study on the Application of Boosting Algorithm for Chl-a Estimation in the Downstream of Nakdong River

Abstract

Objectives:To estimate algae of Chlorophyll-a (Chl-a) with machine learning algorithms, water quality and quantity data of the downstream region of Nakdong River area were used. Methods:At first, the correlation analysis was studied about Chl-a, water quality and quantity data. We have extracted ten important factors for water quality and quantity data about HC (Hapcheon Changnyeong), CH (Changnyeong Haman). Algorithms estimated how ten factors affected Chl-a occurrence each sites. We used algorithms about decision tree, random forest, elastic net, gradient boosting with Python. Results and Discussion:The MSE (Mean of Square Error), RMSE (Root Mean Square Error), R2 (Coefficient of determination) values were used to evaluate excellent algorithms. The gradient boosting showed 56.47 of MSE, 7.51 of RMSE, 0.78 of R2 values for the HC site and 63.82 of MSE, 7.99 of RMSE, 0.76 of R2 values for the CH site. Estimation value for the four algorithms was also verified through the ROC (Receiver Operation Characteristic) curve and AUC (Area Under Curve). As a result of the verification, the AUC value was 0.961 at HC site and the AUC value was 0.885 at CH site. So the gradient boosting algorithm‘s ability to interpret seemed to be excellent. Conclusions:The gradient boosting algorithm showed excellent results for HC and CH sites.

Keywords

• chlorophyll-a
• machine learning
• gradient boosting
• r2
• roc curve