Using SHAP to interpret XGBoost predictions of grassland degradation in Xilingol, China

Abstract

Abstract. Machine learning (ML) and data-driven approaches are increasingly used in many research areas. XGBoost is a tree boosting method that has evolved into a state-of-the-art approach for many ML challenges. However, it has rarely been used in simulations of land use change so far. Xilingol, a typical region for research on serious grassland degradation and its drivers, was selected as a case study to test whether XGBoost can provide alternative insights that conventional land-use models are unable to generate. A set of twenty drivers was analysed using XGBoost, involving four alternative sampling strategies, and SHAP to interpret the results of the purely data-driven approach. The results indicated that, with three of the sampling strategies (over-balanced, balanced and imbalanced), XGBoost achieved similar and robust simulation results. SHAP values were useful for analysing the complex relationship between the different drivers of grassland degradation. Four drivers accounted for 99 % of the grassland degradation dynamics in Xilingol. These four drivers were spatially allocated, and a risk map of further degradation was produced. The limitations of using XGBoost to predict future land-use change are discussed.

Integration of shapley additive explanations with random forest model for quantitative precipitation estimation of mesoscale convective systems

“…Tbb ranks second in the order of feature importance of random forests, and has a clear negative correlation with Shapley values, so the role of tbb in precipitation prediction is further explored by combining feature dependency graphs with interactions. The following analysis is carried out with Frontiers in Environmental Science frontiersin.org reference to existing studies (Wieland et al, 2020;Feng et al, 2021). In Figure 6A, tbb is the first feature and current precipitation is the second feature.…”

Using machine learning to identify patient characteristics to predict mortality of in-patients with COVID-19 in South Florida

“…Beeswarm plot ( Figure 4A ) ( 54 , 55 ), indicated the range across the SHAP value and pointed out the degradation probability, expressed as the logarithm of the odds ( 56 ). We could get a general idea of the directional impact of the features in relation to the distribution of “red” and “blue” dots.…”

“…Each dot on the plot represented a single observation, vertically jittered when too close to each other. Figure 4A also revealed that “mortality” included both linear-dominated relationships (in a box, Figure 4A ), such as diarrhea, sex, heart failure, etc., and non-linear-dominated relations, such as age, BMI, race, etc ( 56 ).…”

A Machine Learning and Computer Vision Study of the Environmental Characteristics of Streetscapes That Affect Pedestrian Satisfaction

“…Ding et al [28] studied the importance of the characteristics of the construction environment for driving distance using the gradient boosting model, which is one of the machine learning models. Moreover, researchers have used SHAP to validate the predictive performance of machine learning-derived land-use change predictions [29], home price prediction through a street view analysis [30], and the prediction of traffic accidents [31]. Thus, research is expanding not only by using machine learning to analyze data but also by using XAI techniques to interpret the results from machine-learning models.…”