Research Article

Multilabel Classification Using Low-Rank Decomposition

Table 3

Performance of each multilabel algorithm (mean ± std. deviation) on the large-scale datasets.

Comparing algorithmsSlashdotrcv1subset1rcv1subset2rcv1subset3rcv1subset4rcv1subset5

One-error ↓
LLRD0.363±0.0260.414 ± 0.0130.411 ± 0.0170.416 ± 0.0290.317±0.0150.401 ± 0.018
MLFE0.374 ± 0.0270.406 ± 0.0180.399 ± 0.0130.402 ± 0.0250.328 ± 0.0130.392 ± 0.008
LIFT0.393 ± 0.0330.427 ± 0.0110.434 ± 0.0170.441 ± 0.0200.363 ± 0.0190.430 ± 0.019
RELIAB0.508 ± 0.0220.449 ± 0.0150.458 ± 0.0280.454 ± 0.0120.433 ± 0.0240.423 ± 0.009
ML20.370 ± 0.0250.404±0.0170.395±0.0180.398±0.0210.323 ± 0.0210.388±0.010
CLR0.965 ± 0.0130.513 ± 0.0220.515 ± 0.0090.518 ± 0.0280.472 ± 0.0310.521 ± 0.021
RAKEL0.602 ± 0.0090.605 ± 0.0130.574 ± 0.0120.585 ± 0.0220.561 ± 0.0220.614 ± 0.009
Coverage ↓
LLRD0.107 ± 0.0100.125±0.0080.121±0.0090.123±0.0060.092 ± 0.0040.116±0.009
MLFE0.126 ± 0.0130.136 ± 0.0050.130 ± 0.0100.129 ± 0.0070.094 ± 0.0070.124 ± 0.007
LIFT0.112 ± 0.0080.144 ± 0.0200.135 ± 0.0080.156 ± 0.0080.113 ± 0.0120.148 ± 0.013
RELIAB0.131 ± 0.0070.152 ± 0.0120.128 ± 0.0140.144 ± 0.0110.105 ± 0.0200.131 ± 0.014
ML20.103±0.0110.138 ± 0.0080.132 ± 0.0100.126 ± 0.0060.078±0.0060.129 ± 0.009
CLR0.254 ± 0.0030.146 ± 0.0180.141 ± 0.0070.137 ± 0.0100.109 ± 0.0180.136 ± 0.011
RAKEL0.226 ± 0.0200.426 ± 0.0230.372 ± 0.0160.381 ± 0.0140.365 ± 0.0090.388 ± 0.020
Ranking loss ↓
LLRD0.090±0.0100.049±0.0040.050±0.0040.052±0.0020.038 ± 0.0020.047±0.003
MLFE0.107 ± 0.0130.052 ± 0.0020.055 ± 0.0070.055 ± 0.0020.040 ± 0.0040.050 ± 0.003
LIFT0.098 ± 0.0160.058 ± 0.0070.057 ± 0.0090.068 ± 0.0040.059 ± 0.0100.055 ± 0.007
RELIAB0.124 ± 0.0030.066 ± 0.0100.063 ± 0.0080.062 ± 0.0040.052 ± 0.0060.063 ± 0.005
ML20.103 ± 0.0120.056 ± 0.0040.057 ± 0.0040.056 ± 0.0030.031±0.0030.050 ± 0.004
CLR0.237 ± 0.0080.062 ± 0.0110.066 ± 0.0080.065 ± 0.0120.047 ± 0.0060.071 ± 0.005
RAKEL0.211 ± 0.0190.226 ± 0.0190.215 ± 0.0170.230 ± 0.0150.235 ± 0.0140.214 ± 0.016
Average precision ↑
LLRD0.725±0.0190.611 ± 0.0100.638 ± 0.0110.634 ± 0.0170.717±0.0080.643 ± 0.011
MLFE0.712 ± 0.0210.618 ± 0.0160.645 ± 0.0090.639 ± 0.0140.708 ± 0.0120.647 ± 0.012
LIFT0.703 ± 0.0100.586 ± 0.0090.598 ± 0.0120.595 ± 0.0110.674 ± 0.0130.598 ± 0.011
RELIAB0.624 ± 0.0140.578 ± 0.0210.611 ± 0.0110.614 ± 0.0180.655 ± 0.0180.604 ± 0.009
ML20.715 ± 0.0220.621±0.0120.647±0.0130.643±0.0160.717±0.0130.650±0.010
CLR0.269 ± 0.0020.575 ± 0.0130.584 ± 0.0210.571 ± 0.0320.614 ± 0.0200.588 ± 0.013
RAKEL0.522 ± 0.0200.395 ± 0.0120.445 ± 0.0180.431 ± 0.0140.450 ± 0.0120.437 ± 0.016
Macroaveraging F1 ↑
LLRD0.427 ± 0.0350.235 ± 0.0200.259 ± 0.0190.213 ± 0.0310.300 ± 0.0190.211 ± 0.020
MLFE0.466 ± 0.0350.198 ± 0.0170.195 ± 0.0560.202 ± 0.0300.249 ± 0.0210.204 ± 0.021
LIFT0.429 ± 0.0370.223 ± 0.0250.186 ± 0.0240.200 ± 0.0310.238 ± 0.0130.196 ± 0.031
RELIAB0.425 ± 0.0290.342±0.0220.338±0.0160.348±0.0140.342±0.0280.352±0.014
ML20.472±0.0290.216 ± 0.0200.206 ± 0.0240.195 ± 0.0300.244 ± 0.0230.208 ± 0.011
CLR0.174 ± 0.0320.285 ± 0.0320.264 ± 0.0210.272 ± 0.0220.311 ± 0.0310.305 ± 0.017
RAKEL0.354 ± 0.0370.269 ± 0.0300.251 ± 0.0140.255 ± 0.0140.263 ± 0.0140.274 ± 0.018
Microaveraging F1 ↑
LLRD0.496 ± 0.0210.393 ± 0.0130.381 ± 0.0170.406 ± 0.0270.470 ± 0.0130.402 ± 0.018
MLFE0.545 ± 0.0190.373 ± 0.0140.375 ± 0.0310.392 ± 0.0240.403 ± 0.0200.381 ± 0.017
LIFT0.510 ± 0.0300.320 ± 0.0170.353 ± 0.0140.347 ± 0.0180.342 ± 0.0240.363 ± 0.008
RELIAB0.453 ± 0.0110.408±0.0100.449±0.0080.451±0.0210.478±0.0160.454±0.012
ML20.556±0.0220.371 ± 0.0140.391 ± 0.0100.383 ± 0.0260.393 ± 0.0220.410 ± 0.015
CLR0.104 ± 0.0050.367 ± 0.0110.368 ± 0.0240.320 ± 0.0240.381 ± 0.0150.372 ± 0.008
RAKEL0.365 ± 0.0200.359 ± 0.0230.348 ± 0.0160.341 ± 0.0160.371 ± 0.0150.342 ± 0.006