File size: 1,596 Bytes
06597ca | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 | B-MO IRON ORE TRAINING SUMMARY
Deposits used: 975 Background points: 1950 Spatial blocks: 122
Model ranking (spatial CV, all points), best first:
0.858 XGBoost [+terrain] finds 60% of deposits at 10% false alarms
0.853 Gradient Boosting [+terrain] finds 59% of deposits at 10% false alarms
0.853 Random Forest [+terrain] finds 59% of deposits at 10% false alarms
0.852 Extra Trees [+terrain] finds 57% of deposits at 10% false alarms
0.849 LightGBM [+terrain] finds 61% of deposits at 10% false alarms
0.848 Logistic Regression [+terrain] finds 60% of deposits at 10% false alarms
0.845 SVM (RBF) [+terrain] finds 61% of deposits at 10% false alarms
0.840 k-Nearest Neighbours [+terrain] finds 57% of deposits at 10% false alarms
0.828 Extra Trees [+thermal] finds 52% of deposits at 10% false alarms
0.823 Random Forest [+thermal] finds 52% of deposits at 10% false alarms
Statistical engine chosen: Statistical engine (likelihood variant), feature set '+terrain', AUC 0.782
Average engine score -> known deposits: 55.4 background: 40.2
Where deposits and background land (out-of-fold, %):
class deposits_pct background_pct
Very low 10.8 50.0
Low 29.0 25.0
Moderate 18.6 15.0
High 18.2 7.0
Very high 23.5 3.0
Top features: B8 (AUC 0.22), brightness (AUC 0.25), B11 (AUC 0.26), slope (AUC 0.73), B3 (AUC 0.29), B12 (AUC 0.30)
Hyperspectral: evaluated on the small covered subset - see model_comparison.csv |