Official-test comparison

HUGIML in the TabArena reference landscape

A dedicated leaderboard view for aligned outer-test comparisons. Search budgets and ensemble construction remain method-specific, so each pool is presented separately.

32 completed datasetsExact outer-split matchingDataset-balanced aggregation200 Elo bootstraps

Dataset-scale analysis

Tuned-model AUC gaps

Models follow their within-quadrant Elo rank. Bars show mean ROC-AUC delta; dots show median delta.

HUGIML mean deltaOfficial tuned model mean deltaMedian delta

Ordering is by TabArena rank, which uses ROC-AUC error for binary datasets and log loss for multiclass datasets. The displayed order is therefore not a ranking by the ROC-AUC values shown at right.

Reference pool

Default · overall

HUGIML plus 16 official variants · 32 completed datasets · Elo anchor RF (default) = 1000

HUGIML position11 / 17within this recomputed pool

Elo1158.695% interval 1067.5–1280.5

Average rank10.00lower is better

Normalized score0.138best = 1, dataset median = 0

Improvability18.41% (8.11%)mean (median); lower is better

Pool-relative results

Leaderboard

Every displayed statistic is recalculated using only this tab's methods.

#	Method	Elo	95% interval	Normalized score	Average rank	Improvability · mean (median)
1	TABICL_GPU (default)	1570.2	1480.0–1692.8	0.689	3.22	4.59% (1.43%)
2	CAT (default)	1487.3	1392.2–1612.5	0.520	4.28	9.12% (3.81%)
3	TABM_GPU (default)	1421.3	1306.9–1562.1	0.457	5.28	10.67% (3.38%)
4	TABPFNV2_GPU (default)	1388.2	1251.3–1541.5	0.528	5.83	8.56% (3.92%)
5	MNCA_GPU (default)	1349.4	1267.4–1452.3	0.304	6.50	11.69% (5.45%)
6	REALMLP (default)	1304.3	1214.6–1409.8	0.220	7.31	12.00% (7.24%)
7	EBM (default)	1295.8	1207.8–1406.8	0.261	7.47	13.50% (6.39%)
8	XGB (default)	1270.4	1175.4–1380.5	0.227	7.94	13.06% (9.65%)
9	TABDPT_GPU (default)	1246.8	1165.1–1345.0	0.245	8.38	12.54% (5.86%)
10	GBM (default)	1233.4	1144.6–1348.2	0.162	8.62	13.88% (9.52%)
11	HUGIML	1158.6	1067.5–1280.5	0.138	10.00	18.41% (8.11%)
12	NN_TORCH (default)	1107.3	1004.9–1213.8	0.059	10.91	17.64% (9.32%)
13	FASTAI (default)	1077.7	987.2–1178.0	0.056	11.41	19.34% (14.15%)
14	RF (default)	1000.0	1000.0–1000.0	0.016	12.62	22.30% (14.50%)
15	LR (default)	969.3	848.3–1099.7	0.038	13.06	26.08% (16.35%)
16	XT (default)	924.0	848.3–974.1	0.029	13.66	24.27% (16.75%)
17	KNN (default)	551.5	217.1–697.9	0.000	16.52	55.91% (51.92%)

Dataset-level pairing

HUGIML versus each official method

Positive error reduction favors HUGIML.

Official method	Wins	Losses	Error reduction · mean (median)
KNN (default)	31	1	46.39% (48.28%)
LR (default)	27	5	8.45% (7.90%)
XT (default)	25	7	4.97% (6.64%)
RF (default)	22	10	0.02% (5.09%)
FASTAI (default)	21	11	-3.58% (2.96%)
NN_TORCH (default)	17	15	-8.76% (0.51%)
GBM (default)	11	21	-12.40% (-3.11%)
EBM (default)	7	25	-14.07% (-3.05%)
XGB (default)	9	23	-15.17% (-3.59%)
TABM_GPU (default)	3	29	-16.72% (-5.54%)
REALMLP (default)	14	18	-19.92% (-1.42%)
MNCA_GPU (default)	6	26	-20.68% (-3.13%)
CAT (default)	5	27	-21.11% (-5.33%)
TABDPT_GPU (default)	13	19	-24.68% (-1.55%)
TABPFNV2_GPU (default)	8	24	-33.07% (-3.66%)
TABICL_GPU (default)	5	27	-40.91% (-4.99%)

Reference pool

Default · binary

HUGIML plus 16 official variants · 26 completed datasets · Elo anchor RF (default) = 1000

HUGIML position11 / 17within this recomputed pool

Elo1152.795% interval 1014.2–1308.7

Average rank10.12lower is better

Normalized score0.158best = 1, dataset median = 0

Improvability16.73% (5.44%)mean (median); lower is better

Pool-relative results

Leaderboard

Every displayed statistic is recalculated using only this tab's methods.

#	Method	Elo	95% interval	Normalized score	Average rank	Improvability · mean (median)
1	TABICL_GPU (default)	1584.3	1464.8–1758.2	0.733	3.04	2.44% (1.00%)
2	CAT (default)	1483.3	1351.1–1649.7	0.564	4.31	7.49% (2.57%)
3	TABM_GPU (default)	1399.0	1274.8–1573.2	0.443	5.62	9.69% (2.90%)
4	MNCA_GPU (default)	1352.2	1264.9–1459.8	0.346	6.42	9.63% (4.73%)
5	TABPFNV2_GPU (default)	1340.3	1201.8–1523.7	0.475	6.63	9.83% (4.61%)
6	EBM (default)	1335.0	1225.0–1474.8	0.305	6.73	10.72% (5.00%)
7	REALMLP (default)	1320.1	1210.0–1431.1	0.260	7.00	9.77% (6.83%)
8	XGB (default)	1245.7	1135.6–1377.0	0.241	8.38	11.94% (8.04%)
9	TABDPT_GPU (default)	1241.6	1130.7–1354.3	0.224	8.46	9.75% (5.86%)
10	GBM (default)	1229.3	1126.7–1355.3	0.160	8.69	12.44% (9.41%)
11	HUGIML	1152.7	1014.2–1308.7	0.158	10.12	16.73% (5.44%)
12	NN_TORCH (default)	1109.5	991.7–1213.4	0.066	10.88	15.41% (8.84%)
13	FASTAI (default)	1091.7	981.9–1191.3	0.052	11.19	16.85% (12.91%)
14	LR (default)	1005.2	844.3–1136.9	0.047	12.58	22.96% (13.29%)
15	RF (default)	1000.0	1000.0–1000.0	0.016	12.65	19.79% (14.26%)
16	XT (default)	926.3	855.5–991.6	0.031	13.65	21.51% (16.29%)
17	KNN (default)	506.1	-3464.0–674.9	0.000	16.63	52.89% (48.49%)

Dataset-level pairing

HUGIML versus each official method

Positive error reduction favors HUGIML.

Official method	Wins	Losses	Error reduction · mean (median)
KNN (default)	25	1	42.79% (44.54%)
LR (default)	21	5	6.23% (4.86%)
XT (default)	21	5	4.32% (6.64%)
RF (default)	18	8	0.50% (4.62%)
FASTAI (default)	16	10	-5.78% (2.20%)
NN_TORCH (default)	12	14	-8.08% (-0.51%)
GBM (default)	9	17	-12.17% (-1.81%)
XGB (default)	8	18	-13.10% (-3.39%)
TABM_GPU (default)	3	23	-13.73% (-3.17%)
EBM (default)	4	22	-14.61% (-3.05%)
CAT (default)	5	21	-20.09% (-4.73%)
REALMLP (default)	11	15	-20.17% (-1.42%)
MNCA_GPU (default)	4	22	-21.16% (-3.56%)
TABPFNV2_GPU (default)	8	18	-21.80% (-3.36%)
TABDPT_GPU (default)	10	16	-28.47% (-1.55%)
TABICL_GPU (default)	4	22	-41.15% (-4.99%)

Reference pool

Default · multiclass

HUGIML plus 16 official variants · 6 completed datasets · Elo anchor RF (default) = 1000

HUGIML position10 / 17within this recomputed pool

Elo1205.695% interval 887.2–1874.7

Average rank9.50lower is better

Normalized score0.051best = 1, dataset median = 0

Improvability25.67% (14.91%)mean (median); lower is better

Pool-relative results

Leaderboard

Every displayed statistic is recalculated using only this tab's methods.

#	Method	Elo	95% interval	Normalized score	Average rank	Improvability · mean (median)
1	TABPFNV2_GPU (default)	1765.1	1592.8–5838.4	0.762	2.33	3.06% (0.19%)
2	TABM_GPU (default)	1606.1	1329.7–2460.6	0.518	3.83	14.93% (5.42%)
3	TABICL_GPU (default)	1591.3	1408.4–2100.6	0.500	4.00	13.92% (4.90%)
4	CAT (default)	1577.0	1354.4–2377.1	0.327	4.17	16.18% (7.35%)
5	XGB (default)	1436.4	1340.0–1913.5	0.163	6.00	17.93% (11.18%)
6	MNCA_GPU (default)	1379.1	1239.9–1918.9	0.121	6.83	20.59% (11.58%)
7	TABDPT_GPU (default)	1302.2	1112.3–1771.2	0.333	8.00	24.64% (7.45%)
8	GBM (default)	1280.6	1018.4–1797.2	0.168	8.33	20.12% (12.26%)
9	REALMLP (default)	1259.2	1103.2–1661.6	0.049	8.67	21.68% (13.77%)
10	HUGIML	1205.6	887.2–1874.7	0.051	9.50	25.67% (14.91%)
11	EBM (default)	1129.2	886.0–1688.4	0.067	10.67	25.53% (19.33%)
12	NN_TORCH (default)	1106.8	918.3–1518.7	0.031	11.00	27.29% (26.63%)
13	FASTAI (default)	1012.5	554.6–1619.9	0.071	12.33	30.11% (23.15%)
14	RF (default)	1000.0	1000.0–1000.0	0.019	12.50	33.20% (21.24%)
15	XT (default)	905.4	735.3–975.3	0.021	13.67	36.22% (21.94%)
16	LR (default)	753.8	73.9–1131.8	0.000	15.17	39.61% (36.47%)
17	KNN (default)	638.8	-3731.2–910.4	0.000	16.00	68.96% (71.60%)

Dataset-level pairing

HUGIML versus each official method

Positive error reduction favors HUGIML.

Official method	Wins	Losses	Error reduction · mean (median)
KNN (default)	6	0	62.00% (66.27%)
LR (default)	6	0	18.05% (22.53%)
XT (default)	4	2	7.80% (8.01%)
FASTAI (default)	5	1	5.98% (5.79%)
RF (default)	4	2	-2.03% (6.74%)
TABDPT_GPU (default)	3	3	-8.22% (-8.37%)
NN_TORCH (default)	5	1	-11.69% (3.48%)
EBM (default)	3	3	-11.70% (-1.77%)
GBM (default)	2	4	-13.44% (-9.16%)
MNCA_GPU (default)	2	4	-18.60% (-2.36%)
REALMLP (default)	3	3	-18.84% (-2.35%)
XGB (default)	1	5	-24.16% (-4.38%)
CAT (default)	0	6	-25.54% (-9.85%)
TABM_GPU (default)	0	6	-29.69% (-11.14%)
TABICL_GPU (default)	1	5	-39.85% (-9.74%)
TABPFNV2_GPU (default)	0	6	-81.94% (-17.30%)

Reference pool

Tuned · overall

HUGIML plus 14 official variants · 32 completed datasets · Elo anchor RF (tuned) = 1000

HUGIML position12 / 15within this recomputed pool

Elo1080.295% interval 967.5–1201.2

Average rank9.81lower is better

Normalized score0.107best = 1, dataset median = 0

Improvability17.60% (9.28%)mean (median); lower is better

Pool-relative results

Leaderboard

Every displayed statistic is recalculated using only this tab's methods.

#	Method	Elo	95% interval	Normalized score	Average rank	Improvability · mean (median)
1	TABM_GPU (tuned)	1427.6	1319.6–1564.7	0.527	4.19	5.98% (2.08%)
2	GBM (tuned)	1395.1	1307.3–1513.1	0.443	4.66	7.64% (2.70%)
3	CAT (tuned)	1378.4	1301.9–1488.1	0.464	4.91	6.49% (3.59%)
4	TABPFNV2_GPU (tuned)	1355.3	1167.7–1504.9	0.572	5.27	5.23% (2.19%)
5	MNCA_GPU (tuned)	1321.2	1239.7–1410.8	0.339	5.81	6.61% (4.19%)
6	XGB (tuned)	1311.7	1235.1–1417.4	0.357	5.97	8.49% (4.54%)
7	REALMLP (tuned)	1278.0	1178.5–1379.6	0.224	6.53	8.05% (5.41%)
8	EBM (tuned)	1211.8	1104.7–1329.4	0.148	7.66	11.69% (5.91%)
9	FASTAI (tuned)	1159.9	1052.7–1294.3	0.147	8.53	12.56% (7.07%)
10	NN_TORCH (tuned)	1119.9	1016.8–1231.5	0.113	9.19	11.09% (6.44%)
11	XT (tuned)	1090.3	1020.0–1166.5	0.112	9.66	14.52% (8.94%)
12	HUGIML	1080.2	967.5–1201.2	0.107	9.81	17.60% (9.28%)
13	RF (tuned)	1000.0	1000.0–1000.0	0.052	10.97	15.54% (9.05%)
14	LR (tuned)	901.3	729.9–1035.3	0.046	12.16	23.74% (15.37%)
15	KNN (tuned)	481.3	81.8–613.9	0.000	14.70	46.64% (40.89%)

Dataset-level pairing

HUGIML versus each official method

Positive error reduction favors HUGIML.

Official method	Wins	Losses	Error reduction · mean (median)
KNN (tuned)	30	2	36.26% (33.98%)
LR (tuned)	27	5	4.55% (5.15%)
RF (tuned)	17	15	-9.22% (0.36%)
XT (tuned)	14	18	-9.37% (-0.69%)
FASTAI (tuned)	13	19	-12.29% (-0.79%)
EBM (tuned)	8	24	-16.66% (-3.36%)
NN_TORCH (tuned)	13	19	-18.29% (-1.61%)
XGB (tuned)	5	27	-21.00% (-5.73%)
GBM (tuned)	4	28	-21.74% (-5.20%)
REALMLP (tuned)	9	23	-24.37% (-3.67%)
CAT (tuned)	5	27	-26.63% (-5.02%)
TABM_GPU (tuned)	4	28	-28.45% (-6.54%)
MNCA_GPU (tuned)	8	24	-28.48% (-5.41%)
TABPFNV2_GPU (tuned)	9	23	-36.83% (-7.52%)

Reference pool

Tuned · binary

HUGIML plus 14 official variants · 26 completed datasets · Elo anchor RF (tuned) = 1000

HUGIML position11 / 15within this recomputed pool

Elo1101.995% interval 957.4–1266.1

Average rank9.54lower is better

Normalized score0.131best = 1, dataset median = 0

Improvability16.04% (7.37%)mean (median); lower is better

Pool-relative results

Leaderboard

Every displayed statistic is recalculated using only this tab's methods.

#	Method	Elo	95% interval	Normalized score	Average rank	Improvability · mean (median)
1	TABM_GPU (tuned)	1434.6	1322.1–1578.7	0.547	4.12	5.43% (1.52%)
2	GBM (tuned)	1410.2	1333.0–1519.8	0.478	4.46	7.27% (2.21%)
3	CAT (tuned)	1381.8	1289.0–1513.8	0.481	4.88	6.87% (2.67%)
4	MNCA_GPU (tuned)	1340.2	1243.0–1461.5	0.375	5.54	5.65% (3.70%)
5	TABPFNV2_GPU (tuned)	1315.5	1168.5–1545.3	0.514	5.94	5.88% (2.56%)
6	XGB (tuned)	1305.1	1218.2–1407.2	0.357	6.12	8.36% (4.45%)
7	REALMLP (tuned)	1253.0	1155.7–1380.6	0.190	7.00	8.08% (5.41%)
8	EBM (tuned)	1241.8	1128.1–1378.2	0.168	7.19	9.79% (5.49%)
9	FASTAI (tuned)	1160.7	1052.8–1296.8	0.134	8.58	11.20% (6.77%)
10	NN_TORCH (tuned)	1128.2	1016.4–1260.2	0.122	9.12	9.15% (6.30%)
11	HUGIML	1101.9	957.4–1266.1	0.131	9.54	16.04% (7.37%)
12	XT (tuned)	1082.3	1002.8–1175.1	0.101	9.85	14.14% (7.63%)
13	RF (tuned)	1000.0	1000.0–1000.0	0.050	11.04	15.04% (7.89%)
14	LR (tuned)	928.6	765.1–1085.4	0.056	11.92	21.20% (12.67%)
15	KNN (tuned)	485.3	160.1–639.6	0.000	14.71	44.69% (38.88%)

Dataset-level pairing

HUGIML versus each official method

Positive error reduction favors HUGIML.

Official method	Wins	Losses	Error reduction · mean (median)
KNN (tuned)	24	2	33.94% (33.98%)
LR (tuned)	22	4	4.64% (4.02%)
XT (tuned)	11	15	-4.11% (-0.69%)
RF (tuned)	14	12	-4.52% (0.36%)
FASTAI (tuned)	12	14	-12.36% (-0.48%)
EBM (tuned)	6	20	-15.96% (-3.04%)
XGB (tuned)	5	21	-17.23% (-3.56%)
GBM (tuned)	4	22	-18.59% (-4.36%)
NN_TORCH (tuned)	11	15	-19.64% (-1.61%)
CAT (tuned)	5	21	-20.24% (-4.37%)
REALMLP (tuned)	8	18	-20.63% (-3.28%)
TABM_GPU (tuned)	4	22	-25.86% (-4.33%)
MNCA_GPU (tuned)	7	19	-27.20% (-5.41%)
TABPFNV2_GPU (tuned)	9	17	-32.52% (-4.80%)

Reference pool

Tuned · multiclass

HUGIML plus 14 official variants · 6 completed datasets · Elo anchor RF (tuned) = 1000

HUGIML position13 / 15within this recomputed pool

Elo972.595% interval 377.3–5444.0

Average rank11.00lower is better

Normalized score0.001best = 1, dataset median = 0

Improvability24.34% (17.10%)mean (median); lower is better

Pool-relative results

Leaderboard

Every displayed statistic is recalculated using only this tab's methods.

#	Method	Elo	95% interval	Normalized score	Average rank	Improvability · mean (median)
1	TABPFNV2_GPU (tuned)	1645.3	1466.3–6492.2	0.821	2.33	2.41% (0.94%)
2	REALMLP (tuned)	1437.4	1169.2–6186.6	0.374	4.50	7.91% (5.49%)
3	TABM_GPU (tuned)	1437.4	1112.8–6542.1	0.443	4.50	8.33% (6.80%)
4	CAT (tuned)	1399.4	1252.3–6213.3	0.391	5.00	4.87% (5.05%)
5	XGB (tuned)	1375.0	1094.8–6286.1	0.355	5.33	9.01% (4.62%)
6	GBM (tuned)	1363.1	970.8–6304.3	0.291	5.50	9.25% (4.89%)
7	MNCA_GPU (tuned)	1259.8	1030.6–5822.8	0.185	7.00	10.76% (6.78%)
8	FASTAI (tuned)	1169.9	559.5–6085.3	0.199	8.33	18.47% (7.88%)
9	XT (tuned)	1135.5	1075.7–1330.0	0.162	8.83	16.15% (12.09%)
10	NN_TORCH (tuned)	1088.4	651.9–5439.0	0.073	9.50	19.50% (12.79%)
11	EBM (tuned)	1076.4	738.5–5846.8	0.062	9.67	19.94% (17.33%)
12	RF (tuned)	1000.0	1000.0–1000.0	0.062	10.67	17.71% (12.99%)
13	HUGIML	972.5	377.3–5444.0	0.001	11.00	24.34% (17.10%)
14	LR (tuned)	742.1	347.8–1095.0	0.000	13.17	34.72% (38.21%)
15	KNN (tuned)	434.6	-3480.0–743.9	0.000	14.67	55.12% (55.22%)

Dataset-level pairing

HUGIML versus each official method

Positive error reduction favors HUGIML.

Official method	Wins	Losses	Error reduction · mean (median)
KNN (tuned)	6	0	46.31% (44.49%)
LR (tuned)	5	1	4.16% (22.67%)
FASTAI (tuned)	1	5	-11.97% (-5.93%)
NN_TORCH (tuned)	2	4	-12.42% (-3.15%)
EBM (tuned)	2	4	-19.72% (-5.33%)
RF (tuned)	3	3	-29.59% (-0.01%)
XT (tuned)	3	3	-32.14% (-1.07%)
MNCA_GPU (tuned)	1	5	-34.06% (-7.22%)
GBM (tuned)	0	6	-35.42% (-14.33%)
XGB (tuned)	0	6	-37.36% (-10.83%)
TABM_GPU (tuned)	0	6	-39.67% (-12.43%)
REALMLP (tuned)	1	5	-40.58% (-13.37%)
CAT (tuned)	0	6	-54.30% (-10.62%)
TABPFNV2_GPU (tuned)	0	6	-55.49% (-15.61%)

Reference pool

Combined · overall

HUGIML plus 45 official variants · 32 completed datasets · Elo anchor RF (tuned) = 1000

HUGIML position35 / 46within this recomputed pool

Elo1056.695% interval 958.8–1181.3

Average rank29.16lower is better

Normalized score0.095best = 1, dataset median = 0

Improvability19.72% (9.40%)mean (median); lower is better

Pool-relative results

Leaderboard

Every displayed statistic is recalculated using only this tab's methods.

#	Method	Elo	95% interval	Normalized score	Average rank	Improvability · mean (median)
1	TABM_GPU (tuned + ensemble)	1481.6	1361.6–1610.5	0.526	9.47	8.36% (3.54%)
2	REALMLP (tuned + ensemble)	1467.1	1368.9–1581.9	0.502	10.00	8.03% (4.44%)
3	TABICL_GPU (default)	1461.3	1378.8–1557.2	0.560	10.22	6.68% (2.96%)
4	AutoGluon_v130_bq_4h8c	1453.1	1350.2–1590.8	0.555	10.53	7.76% (3.28%)
5	GBM (tuned + ensemble)	1423.4	1347.6–1516.1	0.422	11.72	10.19% (4.81%)
6	TABPFNV2_GPU (tuned + ensemble)	1418.1	1230.5–1593.1	0.589	11.94	6.39% (3.02%)
7	TABM_GPU (tuned)	1386.7	1281.9–1509.7	0.433	13.28	9.44% (4.30%)
8	CAT (tuned + ensemble)	1365.7	1289.6–1450.9	0.386	14.22	9.62% (5.79%)
9	GBM (tuned)	1356.1	1278.1–1449.4	0.345	14.66	10.75% (5.12%)
10	CAT (tuned)	1354.7	1283.3–1436.4	0.369	14.72	9.83% (5.65%)
11	CAT (default)	1339.9	1260.0–1436.7	0.345	15.41	10.95% (5.48%)
12	XGB (tuned + ensemble)	1333.9	1256.8–1421.0	0.333	15.69	11.24% (6.11%)
13	TABPFNV2_GPU (tuned)	1333.3	1152.6–1468.7	0.453	15.72	9.28% (7.77%)
14	MNCA_GPU (tuned)	1315.6	1238.3–1400.5	0.310	16.56	10.21% (6.24%)
15	TABPFNV2_GPU (default)	1305.5	1149.9–1465.1	0.418	17.05	10.36% (7.21%)
16	MNCA_GPU (tuned + ensemble)	1294.9	1199.2–1408.6	0.387	17.56	10.81% (6.50%)
17	XGB (tuned)	1294.3	1226.4–1384.9	0.282	17.59	11.55% (6.73%)
18	TABM_GPU (default)	1283.4	1172.9–1402.9	0.294	18.12	12.37% (5.48%)
19	REALMLP (tuned)	1258.1	1170.4–1355.7	0.207	19.38	11.46% (7.26%)
20	EBM (tuned + ensemble)	1255.6	1155.1–1355.1	0.191	19.50	13.90% (7.32%)
21	NN_TORCH (tuned + ensemble)	1252.5	1158.4–1364.2	0.238	19.66	11.81% (6.57%)
22	FASTAI (tuned + ensemble)	1237.4	1142.7–1366.2	0.236	20.41	13.82% (7.96%)
23	MNCA_GPU (default)	1227.4	1155.1–1303.9	0.175	20.91	13.42% (6.73%)
24	EBM (tuned)	1189.3	1089.5–1289.1	0.124	22.81	14.60% (8.17%)
25	EBM (default)	1166.6	1064.0–1265.7	0.144	23.94	15.27% (8.71%)
26	TABDPT_GPU (default)	1159.0	1053.7–1266.3	0.216	24.31	14.13% (6.69%)
27	REALMLP (default)	1152.0	1070.4–1242.6	0.103	24.66	13.79% (9.44%)
28	FASTAI (tuned)	1136.6	1035.6–1251.2	0.109	25.41	15.47% (9.06%)
29	XT (tuned + ensemble)	1120.5	1057.7–1181.0	0.087	26.19	16.26% (8.85%)
30	NN_TORCH (tuned)	1116.5	1021.1–1217.0	0.101	26.38	14.19% (9.46%)
31	XGB (default)	1114.6	1041.0–1191.3	0.095	26.47	14.80% (10.71%)
32	GBM (default)	1069.7	1002.1–1149.8	0.067	28.56	15.57% (10.96%)
33	XT (tuned)	1068.4	1005.8–1128.5	0.063	28.62	17.27% (10.42%)
34	RF (tuned + ensemble)	1057.3	1034.4–1089.3	0.055	29.12	17.02% (10.36%)
35	HUGIML	1056.6	958.8–1181.3	0.095	29.16	19.72% (9.40%)
36	RF (tuned)	1000.0	1000.0–1000.0	0.031	31.59	18.23% (11.79%)
37	NN_TORCH (default)	968.9	881.8–1053.5	0.029	32.84	18.90% (11.65%)
38	FASTAI (default)	950.9	848.8–1059.6	0.031	33.53	20.54% (15.48%)
39	LR (tuned + ensemble)	914.4	789.1–1028.9	0.050	34.86	25.30% (16.59%)
40	LR (tuned)	882.0	745.2–1001.9	0.034	35.95	25.93% (16.71%)
41	LR (default)	868.8	729.1–996.3	0.024	36.38	27.17% (16.75%)
42	RF (default)	865.8	798.2–919.3	0.005	36.47	23.56% (15.51%)
43	XT (default)	803.1	682.3–896.0	0.014	38.28	25.37% (17.66%)
44	KNN (tuned + ensemble)	593.7	407.2–706.5	0.000	42.45	46.03% (40.29%)
45	KNN (tuned)	506.1	265.7–635.7	0.000	43.55	47.87% (43.39%)
46	KNN (default)	303.9	-194.6–485.5	0.000	45.17	56.44% (52.34%)

Dataset-level pairing

HUGIML versus each official method

Positive error reduction favors HUGIML.

Official method	Wins	Losses	Error reduction · mean (median)
KNN (default)	31	1	46.39% (48.28%)
KNN (tuned)	30	2	36.26% (33.98%)
KNN (tuned + ensemble)	29	3	33.47% (32.96%)
LR (default)	27	5	8.45% (7.90%)
XT (default)	25	7	4.97% (6.64%)
LR (tuned)	27	5	4.55% (5.15%)
LR (tuned + ensemble)	27	5	2.86% (4.93%)
RF (default)	22	10	0.02% (5.09%)
FASTAI (default)	21	11	-3.58% (2.96%)
NN_TORCH (default)	17	15	-8.76% (0.51%)
RF (tuned)	17	15	-9.22% (0.36%)
XT (tuned)	14	18	-9.37% (-0.69%)
XT (tuned + ensemble)	12	20	-11.12% (-0.87%)
FASTAI (tuned)	13	19	-12.29% (-0.79%)
GBM (default)	11	21	-12.40% (-3.11%)
RF (tuned + ensemble)	17	15	-12.77% (0.14%)
EBM (default)	7	25	-14.07% (-3.05%)
FASTAI (tuned + ensemble)	10	22	-14.73% (-2.47%)
XGB (default)	9	23	-15.17% (-3.59%)
EBM (tuned)	8	24	-16.66% (-3.36%)
TABM_GPU (default)	3	29	-16.72% (-5.54%)
EBM (tuned + ensemble)	6	26	-18.10% (-4.25%)
NN_TORCH (tuned)	13	19	-18.29% (-1.61%)
REALMLP (default)	14	18	-19.92% (-1.42%)
MNCA_GPU (default)	6	26	-20.68% (-3.13%)
XGB (tuned)	5	27	-21.00% (-5.73%)
CAT (default)	5	27	-21.11% (-5.33%)
XGB (tuned + ensemble)	5	27	-21.34% (-5.80%)
GBM (tuned)	4	28	-21.74% (-5.20%)
GBM (tuned + ensemble)	4	28	-22.49% (-5.69%)
NN_TORCH (tuned + ensemble)	8	24	-23.12% (-3.99%)
REALMLP (tuned)	9	23	-24.37% (-3.67%)
TABDPT_GPU (default)	13	19	-24.68% (-1.55%)
CAT (tuned)	5	27	-26.63% (-5.02%)
CAT (tuned + ensemble)	5	27	-26.88% (-5.14%)
TABM_GPU (tuned)	4	28	-28.45% (-6.54%)
MNCA_GPU (tuned)	8	24	-28.48% (-5.41%)
MNCA_GPU (tuned + ensemble)	7	25	-28.68% (-7.07%)
TABM_GPU (tuned + ensemble)	4	28	-30.67% (-7.77%)
AutoGluon_v130_bq_4h8c	5	27	-31.41% (-6.04%)
REALMLP (tuned + ensemble)	2	30	-31.50% (-6.52%)
TABPFNV2_GPU (default)	8	24	-33.07% (-3.66%)
TABPFNV2_GPU (tuned)	9	23	-36.83% (-7.52%)
TABICL_GPU (default)	5	27	-40.91% (-4.99%)
TABPFNV2_GPU (tuned + ensemble)	8	24	-44.02% (-10.01%)

Reference pool

Combined · binary

HUGIML plus 45 official variants · 26 completed datasets · Elo anchor RF (tuned) = 1000

HUGIML position32 / 46within this recomputed pool

Elo1074.995% interval 943.8–1210.2

Average rank28.62lower is better

Normalized score0.115best = 1, dataset median = 0

Improvability17.96% (8.34%)mean (median); lower is better

Pool-relative results

Leaderboard

Every displayed statistic is recalculated using only this tab's methods.

#	Method	Elo	95% interval	Normalized score	Average rank	Improvability · mean (median)
1	TABICL_GPU (default)	1495.6	1396.8–1605.2	0.602	9.08	4.64% (2.20%)
2	TABM_GPU (tuned + ensemble)	1483.7	1371.9–1628.2	0.532	9.50	7.40% (3.21%)
3	AutoGluon_v130_bq_4h8c	1458.0	1324.3–1613.5	0.568	10.46	6.43% (3.23%)
4	REALMLP (tuned + ensemble)	1449.2	1357.2–1586.7	0.471	10.81	7.74% (3.81%)
5	GBM (tuned + ensemble)	1442.4	1364.1–1551.9	0.439	11.08	9.21% (4.44%)
6	TABM_GPU (tuned)	1390.4	1278.8–1527.5	0.434	13.27	8.40% (3.66%)
7	CAT (default)	1374.9	1276.5–1485.5	0.395	13.96	9.32% (4.37%)
8	TABPFNV2_GPU (tuned + ensemble)	1370.7	1209.0–1596.4	0.522	14.15	6.84% (3.49%)
9	GBM (tuned)	1360.6	1286.4–1472.1	0.349	14.62	9.87% (5.12%)
10	CAT (tuned)	1356.4	1269.1–1463.1	0.369	14.81	9.39% (5.30%)
11	CAT (tuned + ensemble)	1354.0	1276.8–1466.8	0.374	14.92	9.42% (5.34%)
12	XGB (tuned + ensemble)	1333.6	1257.0–1442.1	0.333	15.88	10.45% (6.04%)
13	MNCA_GPU (tuned)	1328.7	1234.3–1450.5	0.335	16.12	8.85% (6.15%)
14	TABPFNV2_GPU (tuned)	1295.5	1140.6–1518.2	0.401	17.73	9.31% (8.14%)
15	MNCA_GPU (tuned + ensemble)	1290.1	1177.5–1439.3	0.401	18.00	9.76% (6.89%)
16	EBM (tuned + ensemble)	1289.3	1201.6–1418.3	0.206	18.04	11.57% (6.20%)
17	XGB (tuned)	1281.6	1215.9–1373.9	0.268	18.42	10.86% (6.47%)
18	TABM_GPU (default)	1281.6	1181.8–1418.4	0.286	18.42	11.35% (4.49%)
19	TABPFNV2_GPU (default)	1271.2	1131.3–1484.1	0.383	18.94	11.61% (7.70%)
20	NN_TORCH (tuned + ensemble)	1260.0	1152.5–1389.3	0.245	19.50	10.07% (6.57%)
21	MNCA_GPU (default)	1252.4	1179.3–1344.8	0.204	19.88	11.41% (5.63%)
22	REALMLP (tuned)	1237.1	1146.0–1360.9	0.169	20.65	10.96% (7.26%)
23	FASTAI (tuned + ensemble)	1235.6	1140.4–1350.7	0.209	20.73	12.29% (7.97%)
24	EBM (tuned)	1215.8	1121.3–1340.1	0.131	21.73	12.27% (7.23%)
25	EBM (default)	1214.3	1110.3–1334.2	0.177	21.81	12.52% (6.62%)
26	REALMLP (default)	1181.4	1090.7–1292.4	0.126	23.46	11.59% (7.73%)
27	TABDPT_GPU (default)	1166.0	1045.3–1279.6	0.198	24.23	11.42% (6.69%)
28	FASTAI (tuned)	1133.3	1037.4–1237.1	0.089	25.85	13.94% (9.06%)
29	NN_TORCH (tuned)	1126.9	1021.9–1252.0	0.106	26.15	12.09% (9.46%)
30	XGB (default)	1115.0	1022.0–1211.8	0.114	26.73	13.67% (10.07%)
31	XT (tuned + ensemble)	1110.2	1037.2–1182.0	0.070	26.96	15.43% (8.35%)
32	HUGIML	1074.9	943.8–1210.2	0.115	28.62	17.96% (8.34%)
33	GBM (default)	1074.9	1000.7–1169.3	0.067	28.62	14.10% (10.37%)
34	XT (tuned)	1063.1	997.2–1138.7	0.050	29.15	16.22% (9.03%)
35	RF (tuned + ensemble)	1043.3	1014.5–1076.3	0.029	30.04	16.44% (9.87%)
36	RF (tuned)	1000.0	1000.0–1000.0	0.021	31.88	17.08% (10.17%)
37	NN_TORCH (default)	983.9	883.9–1084.8	0.036	32.54	16.63% (10.42%)
38	FASTAI (default)	972.2	845.6–1075.9	0.037	33.00	17.95% (14.46%)
39	LR (tuned + ensemble)	940.6	808.4–1061.8	0.061	34.19	22.60% (15.78%)
40	LR (tuned)	914.6	778.4–1046.3	0.041	35.12	23.08% (15.45%)
41	LR (default)	907.9	736.0–1035.3	0.030	35.35	24.00% (15.45%)
42	RF (default)	873.7	795.9–944.1	0.006	36.46	21.10% (14.57%)
43	XT (default)	815.2	689.4–922.4	0.017	38.15	22.65% (17.62%)
44	KNN (tuned + ensemble)	555.3	349.2–697.4	0.000	42.94	45.26% (40.29%)
45	KNN (tuned)	499.0	235.1–650.7	0.000	43.56	45.94% (42.20%)
46	KNN (default)	211.3	-390.1–435.9	0.000	45.48	53.37% (48.49%)

Dataset-level pairing

HUGIML versus each official method

Positive error reduction favors HUGIML.

Official method	Wins	Losses	Error reduction · mean (median)
KNN (default)	25	1	42.79% (44.54%)
KNN (tuned)	24	2	33.94% (33.98%)
KNN (tuned + ensemble)	24	2	32.95% (32.96%)
LR (default)	21	5	6.23% (4.86%)
LR (tuned)	22	4	4.64% (4.02%)
XT (default)	21	5	4.32% (6.64%)
LR (tuned + ensemble)	22	4	3.91% (3.61%)
RF (default)	18	8	0.50% (4.62%)
XT (tuned)	11	15	-4.11% (-0.69%)
RF (tuned)	14	12	-4.52% (0.36%)
XT (tuned + ensemble)	10	16	-5.22% (-0.87%)
RF (tuned + ensemble)	14	12	-5.77% (0.14%)
FASTAI (default)	16	10	-5.78% (2.20%)
NN_TORCH (default)	12	14	-8.08% (-0.51%)
GBM (default)	9	17	-12.17% (-1.81%)
FASTAI (tuned)	12	14	-12.36% (-0.48%)
XGB (default)	8	18	-13.10% (-3.39%)
TABM_GPU (default)	3	23	-13.73% (-3.17%)
EBM (default)	4	22	-14.61% (-3.05%)
FASTAI (tuned + ensemble)	9	17	-14.72% (-2.27%)
EBM (tuned)	6	20	-15.96% (-3.04%)
EBM (tuned + ensemble)	4	22	-17.16% (-3.94%)
XGB (tuned)	5	21	-17.23% (-3.56%)
XGB (tuned + ensemble)	5	21	-17.74% (-4.00%)
GBM (tuned)	4	22	-18.59% (-4.36%)
NN_TORCH (tuned)	11	15	-19.64% (-1.61%)
GBM (tuned + ensemble)	4	22	-19.69% (-5.30%)
CAT (tuned + ensemble)	5	21	-19.81% (-4.39%)
CAT (default)	5	21	-20.09% (-4.73%)
REALMLP (default)	11	15	-20.17% (-1.42%)
CAT (tuned)	5	21	-20.24% (-4.37%)
REALMLP (tuned)	8	18	-20.63% (-3.28%)
MNCA_GPU (default)	4	22	-21.16% (-3.56%)
TABPFNV2_GPU (default)	8	18	-21.80% (-3.36%)
NN_TORCH (tuned + ensemble)	8	18	-23.17% (-2.84%)
TABM_GPU (tuned)	4	22	-25.86% (-4.33%)
REALMLP (tuned + ensemble)	2	24	-26.37% (-5.29%)
MNCA_GPU (tuned)	7	19	-27.20% (-5.41%)
MNCA_GPU (tuned + ensemble)	7	19	-27.34% (-7.07%)
TABM_GPU (tuned + ensemble)	4	22	-27.67% (-5.37%)
TABDPT_GPU (default)	10	16	-28.47% (-1.55%)
AutoGluon_v130_bq_4h8c	5	21	-31.09% (-4.77%)
TABPFNV2_GPU (tuned)	9	17	-32.52% (-4.80%)
TABPFNV2_GPU (tuned + ensemble)	8	18	-38.00% (-4.19%)
TABICL_GPU (default)	4	22	-41.15% (-4.99%)

Reference pool

Combined · multiclass

HUGIML plus 45 official variants · 6 completed datasets · Elo anchor RF (tuned) = 1000

HUGIML position35 / 46within this recomputed pool

Elo968.495% interval 375.8–1601.8

Average rank31.50lower is better

Normalized score0.010best = 1, dataset median = 0

Improvability27.35% (19.04%)mean (median); lower is better

Pool-relative results

Leaderboard

Every displayed statistic is recalculated using only this tab's methods.

#	Method	Elo	95% interval	Normalized score	Average rank	Improvability · mean (median)
1	TABPFNV2_GPU (tuned + ensemble)	1919.4	1716.6–6587.1	0.880	2.33	4.45% (0.34%)
2	REALMLP (tuned + ensemble)	1640.8	1457.6–2420.3	0.637	6.50	9.29% (5.25%)
3	TABPFNV2_GPU (tuned)	1619.7	1450.3–2396.3	0.674	7.00	9.15% (4.94%)
4	TABPFNV2_GPU (default)	1551.3	1347.9–2377.0	0.570	8.83	4.99% (5.03%)
5	TABM_GPU (tuned + ensemble)	1534.5	1252.3–2469.1	0.502	9.33	12.52% (8.00%)
6	AutoGluon_v130_bq_4h8c	1487.4	1010.2–2543.4	0.498	10.83	13.52% (4.01%)
7	CAT (tuned + ensemble)	1477.5	1357.9–2169.4	0.443	11.17	10.49% (8.69%)
8	TABM_GPU (tuned)	1416.7	1121.8–2316.4	0.429	13.33	13.97% (8.96%)
9	REALMLP (tuned)	1403.4	1177.8–2171.4	0.370	13.83	13.63% (8.20%)
10	XGB (tuned)	1399.1	1196.0–2240.6	0.342	14.00	14.53% (10.25%)
11	CAT (tuned)	1390.4	1256.9–2027.0	0.367	14.33	11.73% (9.93%)
12	GBM (tuned + ensemble)	1386.1	1037.4–2226.7	0.349	14.50	14.45% (7.63%)
13	GBM (tuned)	1377.5	1066.2–2217.1	0.326	14.83	14.60% (7.99%)
14	XGB (tuned + ensemble)	1377.5	1162.1–2260.8	0.329	14.83	14.69% (10.31%)
15	TABICL_GPU (default)	1369.0	1212.9–1904.7	0.377	15.17	15.52% (5.82%)
16	MNCA_GPU (tuned + ensemble)	1356.4	1133.9–1901.7	0.322	15.67	15.39% (6.50%)
17	TABM_GPU (default)	1327.5	1017.8–2300.2	0.326	16.83	16.79% (10.04%)
18	MNCA_GPU (tuned)	1287.3	1121.4–1753.0	0.202	18.50	16.11% (8.68%)
19	FASTAI (tuned + ensemble)	1275.4	704.5–2342.1	0.353	19.00	20.49% (6.21%)
20	NN_TORCH (tuned + ensemble)	1243.8	814.6–1975.8	0.210	20.33	19.33% (9.80%)
21	CAT (default)	1212.4	805.9–1934.2	0.130	21.67	18.02% (11.10%)
22	XT (tuned + ensemble)	1185.0	1126.0–1338.5	0.159	22.83	19.84% (10.05%)
23	FASTAI (tuned)	1169.2	533.0–2090.9	0.195	23.50	22.13% (8.28%)
24	TABDPT_GPU (default)	1141.6	694.4–1670.4	0.297	24.67	25.87% (8.34%)
25	RF (tuned + ensemble)	1129.6	1082.4–1243.2	0.170	25.17	19.49% (11.80%)
26	MNCA_GPU (default)	1125.6	838.2–1582.6	0.052	25.33	22.15% (12.83%)
27	XGB (default)	1125.6	859.1–1644.0	0.010	25.33	19.67% (14.30%)
28	EBM (tuned + ensemble)	1113.5	830.0–1900.4	0.124	25.83	23.97% (21.33%)
29	XT (tuned)	1101.4	1051.3–1197.4	0.118	26.33	21.78% (12.47%)
30	NN_TORCH (tuned)	1076.8	632.9–1649.1	0.077	27.33	23.30% (14.59%)
31	EBM (tuned)	1072.7	786.2–1840.5	0.096	27.50	24.67% (22.12%)
32	GBM (default)	1051.8	545.1–1620.0	0.067	28.33	21.95% (13.75%)
33	REALMLP (default)	1013.2	697.0–1476.3	0.000	29.83	23.36% (17.34%)
34	RF (tuned)	1000.0	1000.0–1000.0	0.072	30.33	23.22% (14.73%)
35	HUGIML	968.4	375.8–1601.8	0.010	31.50	27.35% (19.04%)
36	EBM (default)	920.8	562.2–1435.4	0.002	33.17	27.16% (22.42%)
37	NN_TORCH (default)	890.7	668.4–1217.1	0.000	34.17	28.73% (28.30%)
38	FASTAI (default)	837.2	199.5–1396.3	0.005	35.83	31.76% (24.67%)
39	RF (default)	814.3	539.6–905.6	0.000	36.50	34.21% (22.50%)
40	LR (tuned + ensemble)	768.8	515.9–1047.5	0.000	37.75	37.01% (39.42%)
41	XT (default)	725.8	260.4–883.7	0.000	38.83	37.16% (23.19%)
42	LR (tuned)	693.7	331.5–994.8	0.000	39.58	38.28% (39.65%)
43	KNN (tuned + ensemble)	659.2	56.0–825.1	0.000	40.33	49.38% (43.39%)
44	LR (default)	634.6	26.1–1010.1	0.000	40.83	40.92% (39.54%)
45	KNN (tuned)	467.8	-5261.5–743.3	0.000	43.50	56.24% (57.22%)
46	KNN (default)	439.9	-10097.7–762.3	0.000	43.83	69.78% (71.79%)

Dataset-level pairing

HUGIML versus each official method

Positive error reduction favors HUGIML.

Official method	Wins	Losses	Error reduction · mean (median)
KNN (default)	6	0	62.00% (66.27%)
KNN (tuned)	6	0	46.31% (44.49%)
KNN (tuned + ensemble)	5	1	35.73% (38.24%)
LR (default)	6	0	18.05% (22.53%)
XT (default)	4	2	7.80% (8.01%)
FASTAI (default)	5	1	5.98% (5.79%)
LR (tuned)	5	1	4.16% (22.67%)
LR (tuned + ensemble)	5	1	-1.68% (22.16%)
RF (default)	4	2	-2.03% (6.74%)
TABDPT_GPU (default)	3	3	-8.22% (-8.37%)
NN_TORCH (default)	5	1	-11.69% (3.48%)
EBM (default)	3	3	-11.70% (-1.77%)
FASTAI (tuned)	1	5	-11.97% (-5.93%)
NN_TORCH (tuned)	2	4	-12.42% (-3.15%)
GBM (default)	2	4	-13.44% (-9.16%)
FASTAI (tuned + ensemble)	1	5	-14.79% (-8.29%)
MNCA_GPU (default)	2	4	-18.60% (-2.36%)
REALMLP (default)	3	3	-18.84% (-2.35%)
EBM (tuned)	2	4	-19.72% (-5.33%)
EBM (tuned + ensemble)	2	4	-22.16% (-6.05%)
NN_TORCH (tuned + ensemble)	0	6	-22.92% (-7.15%)
XGB (default)	1	5	-24.16% (-4.38%)
CAT (default)	0	6	-25.54% (-9.85%)
RF (tuned)	3	3	-29.59% (-0.01%)
TABM_GPU (default)	0	6	-29.69% (-11.14%)
XT (tuned)	3	3	-32.14% (-1.07%)
AutoGluon_v130_bq_4h8c	0	6	-32.84% (-19.93%)
MNCA_GPU (tuned)	1	5	-34.06% (-7.22%)
MNCA_GPU (tuned + ensemble)	0	6	-34.45% (-8.56%)
GBM (tuned + ensemble)	0	6	-34.63% (-15.59%)
GBM (tuned)	0	6	-35.42% (-14.33%)
XT (tuned + ensemble)	2	4	-36.70% (-3.94%)
XGB (tuned + ensemble)	0	6	-36.97% (-10.83%)
XGB (tuned)	0	6	-37.36% (-10.83%)
TABM_GPU (tuned)	0	6	-39.67% (-12.43%)
TABICL_GPU (default)	1	5	-39.85% (-9.74%)
REALMLP (tuned)	1	5	-40.58% (-13.37%)
RF (tuned + ensemble)	3	3	-43.10% (-2.13%)
TABM_GPU (tuned + ensemble)	0	6	-43.69% (-13.62%)
REALMLP (tuned + ensemble)	0	6	-53.71% (-17.04%)
CAT (tuned)	0	6	-54.30% (-10.62%)
TABPFNV2_GPU (tuned)	0	6	-55.49% (-15.61%)
CAT (tuned + ensemble)	0	6	-57.50% (-12.78%)
TABPFNV2_GPU (tuned + ensemble)	0	6	-70.10% (-23.54%)
TABPFNV2_GPU (default)	0	6	-81.94% (-17.30%)

Default · overall

Predictive metrics and HUGIML deltas

Positive delta favors HUGIML. Rows are ordered by mean ROC AUC.

Method	ROC AUC			Balanced accuracy			F1			Brier
Method	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta
TabPFNv2_GPU	0.8717	0.8853	-0.0116	0.7233	0.7283	-0.0208	0.6085	0.6867	-0.0290	0.1132	0.0960	-0.0101
TabICL_GPU	0.8597	0.8586	-0.0172	0.7027	0.7201	-0.0205	0.5740	0.6460	-0.0230	0.1074	0.0911	-0.0095
CatBoost	0.8577	0.8672	-0.0152	0.6963	0.7058	-0.0142	0.5701	0.6420	-0.0191	0.1102	0.0941	-0.0067
TabM_GPU	0.8551	0.8573	-0.0126	0.6978	0.7073	-0.0157	0.5643	0.6488	-0.0133	0.1113	0.0942	-0.0056
ModernNCA_GPU	0.8546	0.8529	-0.0121	0.6887	0.7043	-0.0066	0.5635	0.6352	-0.0126	0.1112	0.0954	-0.0058
RealMLP	0.8528	0.8548	-0.0102	0.7026	0.7090	-0.0204	0.5799	0.6463	-0.0290	0.1134	0.0978	-0.0035
XGBoost	0.8510	0.8475	-0.0085	0.6905	0.7005	-0.0084	0.5613	0.6452	-0.0104	0.1119	0.0956	-0.0051
ExplainableBM	0.8509	0.8467	-0.0083	0.6852	0.6939	-0.0030	0.5593	0.6404	-0.0084	0.1157	0.0933	-0.0012
TabDPT_GPU	0.8508	0.8412	-0.0083	0.6953	0.7016	-0.0132	0.5663	0.6377	-0.0153	0.1124	0.0986	-0.0045
LightGBM	0.8496	0.8396	-0.0071	0.6902	0.6953	-0.0081	0.5610	0.6440	-0.0101	0.1123	0.0957	-0.0047
NeuralNetTorch	0.8445	0.8463	-0.0020	0.6900	0.7095	-0.0079	0.5570	0.6141	-0.0060	0.1178	0.0972	0.0009
HUGIML	0.8425	0.8365	0.0000	0.6821	0.6981	0.0000	0.5509	0.6004	0.0000	0.1169	0.0976	0.0000
RandomForest	0.8391	0.8403	0.0034	0.6867	0.6818	-0.0046	0.5620	0.6067	-0.0111	0.1198	0.1014	0.0029
NeuralNetFastAI	0.8372	0.8290	0.0054	0.6759	0.6921	0.0062	0.5373	0.6049	0.0136	0.1198	0.0941	0.0029
ExtraTrees	0.8345	0.8347	0.0081	0.6726	0.6622	0.0095	0.5411	0.5698	0.0099	0.1233	0.1018	0.0064
LinearModel	0.8255	0.8082	0.0170	0.6529	0.6412	0.0292	0.5085	0.5741	0.0424	0.1300	0.1030	0.0130
KNeighbors	0.6615	0.6288	*0.1696*	0.5655	0.5514	*0.0972*	0.3620	0.3048	*0.1357*	0.1860	0.1742	*0.0633*

Default · binary

Predictive metrics and HUGIML deltas

Positive delta favors HUGIML. Rows are ordered by mean ROC AUC.

Method	ROC AUC			Balanced accuracy			F1			Brier
Method	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta
TabPFNv2_GPU	0.8496	0.8331	-0.0108	0.7106	0.7253	-0.0183	0.5608	0.6479	-0.0302	0.1004	0.0920	-0.0053
TabICL_GPU	0.8397	0.8279	-0.0177	0.6890	0.7101	-0.0194	0.5301	0.6010	-0.0234	0.0958	0.0883	-0.0061
CatBoost	0.8378	0.8234	-0.0158	0.6846	0.6886	-0.0150	0.5282	0.5911	-0.0215	0.0978	0.0893	-0.0041
ModernNCA_GPU	0.8344	0.8148	-0.0124	0.6751	0.6960	-0.0056	0.5196	0.5843	-0.0128	0.0987	0.0902	-0.0033
TabM_GPU	0.8344	0.8247	-0.0124	0.6848	0.6946	-0.0153	0.5200	0.5906	-0.0132	0.0989	0.0897	-0.0030
RealMLP	0.8330	0.8224	-0.0110	0.6905	0.7014	-0.0210	0.5385	0.5833	-0.0317	0.0992	0.0917	-0.0027
ExplainableBM	0.8329	0.8075	-0.0109	0.6798	0.6939	-0.0102	0.5235	0.5781	-0.0167	0.0993	0.0896	-0.0027
XGBoost	0.8300	0.8144	-0.0080	0.6756	0.6719	-0.0061	0.5154	0.5858	-0.0086	0.0997	0.0901	-0.0023
TabDPT_GPU	0.8297	0.8096	-0.0077	0.6837	0.7002	-0.0141	0.5250	0.5809	-0.0183	0.0984	0.0904	-0.0035
LightGBM	0.8295	0.8153	-0.0075	0.6768	0.6776	-0.0072	0.5171	0.5872	-0.0104	0.0995	0.0903	-0.0025
NeuralNetTorch	0.8251	0.8168	-0.0030	0.6800	0.7002	-0.0105	0.5166	0.5797	-0.0099	0.1016	0.0906	-0.0003
HUGIML	0.8220	0.7947	0.0000	0.6696	0.6807	0.0000	0.5067	0.5607	0.0000	0.1019	0.0904	0.0000
NeuralNetFastAI	0.8192	0.8082	0.0028	0.6711	0.6757	-0.0015	0.4986	0.5781	0.0082	0.1015	0.0904	-0.0004
RandomForest	0.8162	0.8016	0.0058	0.6731	0.6642	-0.0036	0.5188	0.5397	-0.0120	0.1051	0.0937	0.0032
ExtraTrees	0.8108	0.7940	0.0112	0.6579	0.6528	0.0117	0.4943	0.5171	0.0124	0.1071	0.0951	0.0051
LinearModel	0.8070	0.7836	0.0150	0.6448	0.6215	0.0247	0.4660	0.4554	0.0407	0.1090	0.0961	0.0071
KNeighbors	0.6451	0.6074	*0.1737*	0.5832	0.5514	*0.0813*	0.3437	0.2606	*0.1269*	0.1541	0.1418	*0.0476*

Default · multiclass

Predictive metrics and HUGIML deltas

Positive delta favors HUGIML. Rows are ordered by mean ROC AUC.

Method	ROC AUC			Balanced accuracy			F1			Brier
Method	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta
TabICL_GPU	0.9466	0.9656	-0.0151	0.7618	0.8696	-0.0253	0.7640	0.8673	-0.0215	0.1577	0.1583	-0.0243
TabPFNv2_GPU	0.9456	0.9600	-0.0141	0.7657	0.8656	-0.0293	0.7673	0.8622	-0.0248	0.1558	0.1574	-0.0262
TabM_GPU	0.9450	0.9485	-0.0135	0.7540	0.8475	-0.0176	0.7562	0.8410	-0.0137	0.1652	0.1571	-0.0168
CatBoost	0.9442	0.9549	-0.0127	0.7471	0.8400	-0.0107	0.7514	0.8416	-0.0090	0.1641	0.1621	-0.0178
TabDPT_GPU	0.9423	0.9643	-0.0108	0.7457	0.8620	-0.0093	0.7451	0.8658	-0.0026	0.1732	0.1693	-0.0088
ModernNCA_GPU	0.9421	0.9588	-0.0106	0.7478	0.8428	-0.0114	0.7539	0.8468	-0.0114	0.1653	0.1673	-0.0167
XGBoost	0.9421	0.9556	-0.0106	0.7551	0.8530	-0.0187	0.7606	0.8504	-0.0182	0.1647	0.1647	-0.0173
RealMLP	0.9385	0.9400	-0.0070	0.7546	0.8257	-0.0182	0.7594	0.8223	-0.0169	0.1749	0.1648	-0.0070
RandomForest	0.9383	0.9556	-0.0069	0.7456	0.8345	-0.0092	0.7495	0.8384	-0.0070	0.1835	0.1753	0.0016
ExtraTrees	0.9370	0.9545	-0.0056	0.7365	0.8264	-0.0001	0.7437	0.8335	-0.0012	0.1937	0.1881	0.0117
LightGBM	0.9369	0.9532	-0.0054	0.7487	0.8437	-0.0122	0.7514	0.8421	-0.0089	0.1677	0.1695	-0.0143
HUGIML	0.9315	0.9348	0.0000	0.7364	0.7965	0.0000	0.7425	0.7979	0.0000	0.1820	0.1668	0.0000
NeuralNetTorch	0.9289	0.9267	0.0025	0.7333	0.7985	0.0031	0.7319	0.7950	0.0106	0.1881	0.1735	0.0061
ExplainableBM	0.9285	0.9194	0.0029	0.7085	0.7126	0.0279	0.7145	0.7174	0.0280	0.1869	0.1923	0.0049
NeuralNetFastAI	0.9150	0.9203	0.0165	0.6970	0.7002	0.0394	0.7053	0.7100	0.0372	0.1994	0.1897	0.0174
LinearModel	0.9060	0.9157	0.0255	0.6878	0.6801	0.0486	0.6925	0.6895	0.0500	0.2207	0.2037	0.0388
KNeighbors	0.7599	0.8084	*0.1448*	0.4593	0.5080	*0.1923*	0.4722	0.5289	*0.1883*	0.3773	0.3734	*0.1576*

Tuned · overall

Predictive metrics and HUGIML deltas

Positive delta favors HUGIML. Rows are ordered by mean ROC AUC.

Method	ROC AUC			Balanced accuracy			F1			Brier
Method	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta
TabPFNv2_GPU	0.8747	0.8913	-0.0145	0.7244	0.7268	-0.0219	0.6145	0.6869	-0.0350	0.1127	0.0970	-0.0106
CatBoost	0.8580	0.8675	-0.0155	0.6986	0.7044	-0.0165	0.5720	0.6418	-0.0211	0.1093	0.0939	-0.0076
TabM_GPU	0.8579	0.8593	-0.0153	0.6987	0.7200	-0.0166	0.5728	0.6483	-0.0219	0.1087	0.0929	-0.0082
LightGBM	0.8573	0.8582	-0.0148	0.6967	0.7076	-0.0146	0.5695	0.6436	-0.0186	0.1091	0.0932	-0.0078
ModernNCA_GPU	0.8573	0.8632	-0.0147	0.7018	0.7101	-0.0197	0.5695	0.6390	-0.0186	0.1091	0.0955	-0.0079
XGBoost	0.8564	0.8624	-0.0139	0.6975	0.7047	-0.0154	0.5707	0.6486	-0.0197	0.1096	0.0938	-0.0074
RealMLP	0.8550	0.8560	-0.0125	0.7035	0.7123	-0.0213	0.5780	0.6476	-0.0270	0.1098	0.0993	-0.0071
ExplainableBM	0.8515	0.8524	-0.0089	0.6869	0.6911	-0.0048	0.5616	0.6454	-0.0107	0.1148	0.0935	-0.0021
NeuralNetTorch	0.8513	0.8497	-0.0087	0.6931	0.7093	-0.0110	0.5613	0.6484	-0.0104	0.1137	0.0978	-0.0033
ExtraTrees	0.8469	0.8393	-0.0043	0.6773	0.6650	0.0048	0.5379	0.5901	0.0131	0.1145	0.0956	-0.0024
NeuralNetFastAI	0.8468	0.8475	-0.0043	0.6824	0.7047	-0.0003	0.5418	0.6159	0.0092	0.1149	0.0932	-0.0020
RandomForest	0.8462	0.8352	-0.0037	0.6918	0.6948	-0.0097	0.5658	0.6312	-0.0149	0.1146	0.0962	-0.0023
HUGIML	0.8425	0.8365	0.0000	0.6821	0.6981	0.0000	0.5509	0.6004	0.0000	0.1169	0.0976	0.0000
LinearModel	0.8280	0.8168	0.0145	0.6564	0.6397	0.0258	0.5132	0.5731	0.0377	0.1284	0.1013	0.0115
KNeighbors	0.7053	0.6901	*0.1257*	0.5506	0.5206	*0.1121*	0.3143	0.2496	*0.1834*	0.1713	0.1484	*0.0487*

Tuned · binary

Predictive metrics and HUGIML deltas

Positive delta favors HUGIML. Rows are ordered by mean ROC AUC.

Method	ROC AUC			Balanced accuracy			F1			Brier
Method	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta
TabPFNv2_GPU	0.8520	0.8309	-0.0133	0.7119	0.7255	-0.0195	0.5691	0.6505	-0.0385	0.1006	0.0940	-0.0050
CatBoost	0.8378	0.8228	-0.0158	0.6848	0.6906	-0.0152	0.5278	0.5877	-0.0211	0.0977	0.0894	-0.0042
TabM_GPU	0.8371	0.8240	-0.0151	0.6843	0.7100	-0.0147	0.5288	0.6012	-0.0220	0.0971	0.0896	-0.0048
ModernNCA_GPU	0.8368	0.8240	-0.0148	0.6889	0.7071	-0.0193	0.5260	0.5915	-0.0193	0.0976	0.0902	-0.0044
LightGBM	0.8368	0.8263	-0.0148	0.6829	0.6971	-0.0133	0.5255	0.5835	-0.0187	0.0975	0.0896	-0.0044
XGBoost	0.8355	0.8191	-0.0135	0.6832	0.6900	-0.0136	0.5262	0.5824	-0.0195	0.0981	0.0895	-0.0038
RealMLP	0.8339	0.8232	-0.0119	0.6888	0.7090	-0.0193	0.5340	0.5915	-0.0273	0.0989	0.0928	-0.0030
ExplainableBM	0.8328	0.8085	-0.0108	0.6811	0.6911	-0.0115	0.5253	0.5822	-0.0186	0.0990	0.0897	-0.0029
NeuralNetTorch	0.8319	0.8191	-0.0099	0.6804	0.6991	-0.0108	0.5184	0.5933	-0.0117	0.0993	0.0912	-0.0026
NeuralNetFastAI	0.8274	0.8182	-0.0054	0.6703	0.6754	-0.0007	0.4969	0.5665	0.0099	0.0994	0.0902	-0.0025
RandomForest	0.8262	0.8099	-0.0042	0.6782	0.6694	-0.0086	0.5221	0.5816	-0.0153	0.1016	0.0913	-0.0004
ExtraTrees	0.8259	0.8100	-0.0039	0.6603	0.6440	0.0092	0.4884	0.4821	0.0184	0.1022	0.0913	0.0002
HUGIML	0.8220	0.7947	0.0000	0.6696	0.6807	0.0000	0.5067	0.5607	0.0000	0.1019	0.0904	0.0000
LinearModel	0.8106	0.7912	0.0114	0.6489	0.6313	0.0207	0.4719	0.4776	0.0348	0.1079	0.0959	0.0059
KNeighbors	0.6945	0.6769	*0.1243*	0.5756	0.5206	*0.0890*	0.3004	0.1884	*0.1702*	0.1385	0.1251	*0.0321*

Tuned · multiclass

Predictive metrics and HUGIML deltas

Positive delta favors HUGIML. Rows are ordered by mean ROC AUC.

Method	ROC AUC			Balanced accuracy			F1			Brier
Method	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta	Mean	Median	HUGIML delta
TabPFNv2_GPU	0.9501	0.9622	-0.0187	0.7660	0.8754	-0.0295	0.7659	0.8689	-0.0234	0.1529	0.1523	-0.0290
TabM_GPU	0.9478	0.9557	-0.0163	0.7612	0.8659	-0.0248	0.7636	0.8569	-0.0211	0.1590	0.1536	-0.0229
XGBoost	0.9471	0.9579	-0.0156	0.7594	0.8529	-0.0230	0.7632	0.8503	-0.0207	0.1592	0.1595	-0.0228
RealMLP	0.9465	0.9580	-0.0150	0.7668	0.8670	-0.0304	0.7686	0.8592	-0.0261	0.1572	0.1544	-0.0248
LightGBM	0.9460	0.9579	-0.0145	0.7565	0.8497	-0.0200	0.7605	0.8489	-0.0180	0.1595	0.1612	-0.0225
CatBoost	0.9458	0.9580	-0.0143	0.7588	0.8480	-0.0224	0.7636	0.8492	-0.0211	0.1595	0.1596	-0.0224
ModernNCA_GPU	0.9457	0.9608	-0.0143	0.7578	0.8646	-0.0214	0.7581	0.8587	-0.0156	0.1589	0.1604	-0.0231
ExtraTrees	0.9379	0.9582	-0.0064	0.7509	0.8483	-0.0144	0.7525	0.8491	-0.0100	0.1679	0.1660	-0.0141
NeuralNetTorch	0.9354	0.9384	-0.0039	0.7486	0.8348	-0.0121	0.7472	0.8267	-0.0047	0.1761	0.1594	-0.0059
RandomForest	0.9330	0.9549	-0.0015	0.7506	0.8434	-0.0141	0.7553	0.8443	-0.0129	0.1711	0.1664	-0.0109
ExplainableBM	0.9325	0.9212	-0.0011	0.7120	0.7163	0.0244	0.7188	0.7214	0.0236	0.1834	0.1888	0.0014
HUGIML	0.9315	0.9348	0.0000	0.7364	0.7965	0.0000	0.7425	0.7979	0.0000	0.1820	0.1668	0.0000
NeuralNetFastAI	0.9308	0.9360	0.0006	0.7348	0.7950	0.0017	0.7362	0.7954	0.0062	0.1821	0.1606	0.0002
LinearModel	0.9036	0.9156	0.0279	0.6887	0.6801	0.0477	0.6923	0.6894	0.0502	0.2176	0.2038	0.0357
KNeighbors	0.7705	0.8044	*0.1342*	0.4006	0.4178	*0.2510*	0.3978	0.4135	*0.2627*	0.3682	0.3794	*0.1484*

Methodology and interpretation

Comparison design

Local HUGIML results are joined to published TabArena results by exact dataset name and outer-split number.
Binary error is 1 − ROC AUC; multiclass error is log loss. Split errors are averaged within dataset before aggregation, giving every dataset equal weight.
Each tab is an independent comparison pool. Elo, ranks, normalized score, improvability, confidence intervals, and HUGIML pairwise results are recomputed within that pool.
AutoMLPipelineFeatureGenerator is fitted separately on every inner child training fold and retained with that child for validation and outer-test transformation; no validation or test rows contribute to preprocessing state.
The outer tests and metrics align. HUGIML evaluates 16 configurations, while each tuned official baseline evaluates 200; retained ensembles and compute budgets remain method-specific.
Detailed metric tables show mean, median, and the paired HUGIML advantage. Positive deltas favor HUGIML; Brier uses the lower-is-better direction.