Table 2.

Details of model training.

Algorithm	#Parameters optimized	#Candidates in random search	Time taken (2 map, 4 map)	#Candidates in grid search	Time taken (2 map, 4 map)
LR	6	28	5 min, 33 min	3	20 min, 53 min
RF	7	80	1 h 53 min, 3 h 1min	81	4 h 26 min, 6 58 min
XGB	5	10	30 min, 57 min	27	1 h 38 min, 2 h 53 min
SVM	3	30	22 h 33 min, N/A	N/A	N/A

Four different algorithms were used to train models: Logistic Regression (LR), Random Forest (RF), XGBoost (XGB) and Support Vector Machine (SVM). Each algorithm has different hyperparameters, and the number of different hyperparameters that were optimized here is shown (“^#Parameters optimized”). Except for SVM, the parameters were optimized by first running a random search, training and testing models with a number of different random hyperparameter models (“^#Candidates in random search”). The best performing combination was used to create the range for a more refined grid search. The number of candidates tested in the grid search was determined by the number of parameters that were optimized and the possible values for each parameter (e.g., whether values were discrete or continuous). The time that was taken to run all the different combinations is included. Six CPUs were used in parallel.