Decision Tree Classifier

Model overview

Decision tree classifier for predicting taxonomy IDs from product tags. Unlike parameter-constrained models, this classifier uses ALL available tags. The tree structure provides natural interpretability - each decision node tests whether a specific tag is present, leading to a taxonomy prediction at the leaves.

5 CV folds

49.42% Mean accuracy

0.5290 Mean F1

0.232188 Mean p-adic loss

120,392 Avg effective params

Total train samples: 92,388
Total test samples: 23,097
Number of tags (input features): 7,726
Number of classes: 991
Avg tree depth: 1876.6
Avg number of nodes: 12,092
Avg number of leaves: 6,046
Effective params formula: nodes × log₂(classes) = 12,092 × 9.95 ≈ 120,392

Cross-validation results

Fold	Accuracy	F1	P-adic loss (mean)	Tree depth	Nodes	Effective params
0	50.54%	0.5429	0.209426	1884	12,195	121,374
1	48.73%	0.5217	0.238152	1904	11,895	118,422
2	49.33%	0.5273	0.239897	1871	12,067	120,117
3	49.19%	0.5269	0.242073	1856	12,117	120,668
4	49.30%	0.5263	0.231392	1868	12,185	121,380

Feature importances (Top 50)

Feature importances are averaged across all 5 folds. The "Folds" column shows how many folds used this feature.

Rank	Tag	Avg Importance	Folds
1	NEW	0.004873	5/5
2	ACCESSORIES	0.003026	5/5
3	NEW ARRIVALS	0.002506	5/5
4	BLACK	0.002484	5/5
5	BIRTHDAY	0.002480	5/5
6	KIDS	0.002073	5/5
7	RETAIL	0.001991	5/5
8	WOMEN	0.001930	5/5
9	MEN	0.001874	5/5
10	WHITE	0.001766	5/5
11	HAT	0.001592	5/5
12	DRESS	0.001552	5/5
13	WOMENS	0.001547	5/5
14	PRINTED SILK	0.001483	1/5
15	PIVIT	0.001480	2/5
16	_RELATED_HOT-CEREAL-LANDING-PAGE	0.001477	3/5
17	MOTORCYCLE RAMP	0.001477	3/5
18	TV SPEAKER	0.001477	4/5
19	INKYBAY_CUSTOM	0.001477	5/5
20	TOWEL-RAILS	0.001477	5/5
21	PLANT STANDS AND BENCHES	0.001477	5/5
22	CHEESE TOOL	0.001477	5/5
23	CAR COASTERS	0.001477	5/5
24	KEYS	0.001477	5/5
25	SIRUP	0.001477	5/5
26	CIDER	0.001477	5/5
27	TIRE SEALANT	0.001477	5/5
28	RAFCAMERA	0.001477	5/5
29	ELECTRIC-SKATEBOARDS	0.001477	5/5
30	PALEO HOT CEREAL	0.001477	1/5
31	PHANTOMX	0.001477	1/5
32	SOUNDBAR	0.001477	1/5
33	BICYCLES AND MOTORCYCLES	0.001477	2/5
34	JIGS & RIGS	0.001477	2/5
35	SSET	0.001477	1/5
36	IMT TOOLS	0.001476	4/5
37	ROAD	0.001476	4/5
38	PRINTED POCKET SQ	0.001476	4/5
39	CUTTER REPLACEMENT	0.001472	1/5
40	TEETHING ANKLET	0.001472	1/5
41	TRAILER JIGS	0.001472	1/5
42	PUBLISHEDNEW	0.001469	1/5
43	1666	0.001464	1/5
44	DUPION	0.001464	1/5
45	CLOTHING	0.001461	5/5
46	OVER-10	0.001460	4/5
47	SAREE	0.001459	4/5
48	45MM	0.001459	1/5
49	HARLEY DAVIDSON MOTORCYCLE RAIN GEAR	0.001452	1/5
50	SMALL WEAPONS	0.001445	5/5