Predicting cash holdings using supervised machine learning algorithms

Financial Innovation

Table 13 Performance metrics comparison for the algorithms

	MLR	KNN	SVR	Decision trees	Random forest	XGBoost
RMSE	0.1036	0.1064	0.0796	0.0899	0.071	0.0599
% Improvement in RMSE	0	− 2.7027	23.1660	13.2239	31.4672	42.1815
R²	0.1626	0.1228	0.5152	0.3822	0.6147	0.7258
% Improvement in R²	32.4104	0	319.5440	211.2378	400.5700	491.0423

The algorithms applied in the study are evaluated based on the RMSE and R² values. For the RMSE, the minimum value is obtained from the XGBoost algorithm. This RMSE is 42.18% lower than that of MLR algorithm, which is the highest. For R², the maximum value is obtained by XGBoost algorithm again. This R² is 42.18% lower than that of KNN algorithm, which is the lowest R² value among all