Distributionally robust chance constrained svm model with $\ell_2$-Wasserstein distance

Qing Ma; Yanjun Wang

doi:10.3934/jimo.2021212

Article Contents

2023, Volume 19, Issue 2: 916-931. Doi: 10.3934/jimo.2021212

This issue Previous Article A new hybrid

$ l_p $

$ l_2 $

model for sparse solutions with applications to image processing Next Article Optimal control of fish-feeding in a three-dimensional calm freshwater pond considering environmental concern

Distributionally robust chance constrained svm model with $\ell_2$-Wasserstein distance

Qing Ma and
Yanjun Wang^,

School of Mathematics, Shanghai University of Finance and Economics, Shanghai 200433, China

^* Corresponding author: wangyanjun@mail.shufe.edu.cn
^* Corresponding author: wangyanjun@mail.shufe.edu.cn

Received: April 2021

Revised: November 2021

Early access: December 2021

Published: February 2023

Abstract / Introduction Full Text(HTML) Figure(0) / Table(4) Related Papers Cited by

Abstract

In this paper, we propose a distributionally robust chance-constrained SVM model with $ \ell_2 $-Wasserstein ambiguity. We present equivalent formulations of distributionally robust chance constraints based on $ \ell_2 $-Wasserstein ambiguity. In terms of this method, the distributionally robust chance-constrained SVM model can be transformed into a solvable linear 0-1 mixed integer programming problem when the $ \ell_2 $-Wasserstein distance is discrete form. The DRCC-SVM model could be transformed into a tractable 0-1 mixed-integer SOCP programming problem for the continuous case. Finally, numerical experiments are given to illustrate the effectiveness and feasibility of our model.

Keywords:

Mathematics Subject Classification: Primary: 58F15, 58F17; Secondary: 53C35.

Citation:

Full Text(HTML)

Table 1. Mean performance comparison

	AUC	(S.E.)
Classical SVM	0.9039	0.0046
$ \ell_2 $-Wasserstein SVM	0.9105	0.0020

| Show Table

DownLoad: CSV

Table 2. UCI Database

Date set	classification	numbers	feature
Sonar	2	208	60
Pima	2	267	22
Heart	2	270	12
BUPA Liver	2	345	6
Ionosphere	2	351	34
Australian	2	690	14
Breast Cancer	2	699	13
Bank	2	4521	16

| Show Table

DownLoad: CSV

Table 3.

		Classic SVM		$ \ell_2 $-Wasserstein SVM		Soft margin SVM
Data set	n	Accuracy	(S.E.)	Accuracy	(S.E.)	Accuracy	(S.E.)
Sonar	50	0.814	0.0064	0.801	0.0050	0.799	0.0049
Sonar	70	0.856	0.0045	0.851	0.0041	0.850	0.0040
Sonar	100	0.872	0.0038	0.862	0.0033	0.860	0.0032
Pima	50	0.732	0.0055	0.723	0.0048	0.722	0.0047
Pima	70	0.745	0.0051	0.741	0.0043	0.741	0.0043
Pima	100	0.751	0.0046	0.743	0.0034	0.743	0.0034
Heart	50	0.791	0.0027	0.788	0.0025	0.787	0.0024
Heart	70	0.821	0.0024	0.815	0.0020	0.814	0.0018
Heart	100	0.833	0.0016	0.831	0.0011	0.827	0.0009
BUPA Liver	50	0.773	0.0035	0.768	0.0033	0.765	0.0028
BUPA Liver	70	0.789	0.0032	0.788	0.0031	0.788	0.0031
BUPA Liver	100	0.801	0.0025	0.801	0.0024	0.799	0.0023
Ionosphere	50	0.812	0.0054	0.810	0.0051	0.808	0.0050
Ionosphere	70	0.834	0.0037	0.842	0.0035	0.832	0.0035
Ionosphere	100	0.852	0.0033	0.848	0.0022	0.845	0.0020
Australian	50	0.812	0.0022	0.810	0.0020	0.807	0.0015
Australian	70	0.823	0.0019	0.832	0.0016	0.822	0.0015
Australian	100	0.833	0.0015	0.830	0.0013	0.830	0.0013
Breast Cancer	50	0.952	0.0023	0.951	0.0022	0.951	0.0022
Breast Cancer	70	0.955	0.0020	0.955	0.0018	0.952	0.0017
Breast Cancer	100	0.964	0.0016	0.969	0.0011	0.952	0.0009

| Show Table

DownLoad: CSV

Table 4.

Data set	n	Classic SVM		L2-Wasserstein SVM		Soft-margin SVM
Data set	n	Accuracy	Accuracy (10% noise)	Accuracy	Accuracy (10% noise)	Accuracy	Accuracy (10% noise)
Bank	100	0.8421	0.7870	0.8841	0.8842	0.8823	0.8743
	200	0.8377	0.7798	0.8844	0.8846	0.8823	0.8773
	300	0.8379	0.7785	0.8846	0.8845	0.8824	0.8768
	400	0.8377	0.7757	0.8850	0.8847	0.8821	0.8763
	500	0.8407	0.7735	0.8853	0.8854	0.8833	0.8765
	600	0.8389	0.7759	0.8855	0.8852	0.8831	0.8775

| Show Table

DownLoad: CSV

Related Papers

Cited by

References

[1]	G. Aurora and M. C. Victoria, Towards energy efficiency smart buildings models based on in telligent data analytics, Procedia Computer Science, 83 (2016), 994-999.
[2]	Y. Q. Bai and K. J. Shen, Alternating direction method of multipliers for $\ell_1$-$\ell_2$ regularized logistic regression model, J. Oper. Res. Soc. China, 4 (2016), 243-253. doi: 10.1007/s40305-015-0090-2.
[3]	Y. Q. Bai, Y. J. Shen and K. J. Shen, Consensus proximal support vector machine for classficication problems with sparse solutions, J. Oper. Res. Soc. China, 2 (2014), 57-79. doi: 10.1007/s40305-014-0037-z.
[4]	C. Bhattacharyya, Robust classification of noisy data using second order cone programming approach, In Intelligent Sensing and Information Processing, IEEE, (2004), 433–438.
[5]	J. Blanchet, L. Chen and X. Y. Zhou, Distributionally robust mean-variance portfolio selection with wasserstein distances, preprint, arXiv: 1802.04885.
[6]	G. C. Calafiore and L. Ghaoui, On distributionally robust chance-constrained linear programs, J. Optim. Theory Appl., 130 (2006), 1-22. doi: 10.1007/s10957-006-9084-x.
[7]	S. Chao, X. Huang and F. You, Data-driven robust optimization based on kernel learning, Computers and Chemical Engineering, 106 (2017), 464-479.
[8]	A. Charnes, W. W. Cooper and G. H. Symonds, Cost horizons and certainty equivalents: An approach to stochastic programming of heating oil, Management Science, 4 (1958), 235-263. doi: 10.1287/mnsc.4.3.235.
[9]	C. Cortes and V. Vapnik, Support-vector networks, Machine Learning, 20 (1995), 273-297. doi: 10.1007/BF00994018.
[10]	M. D. Dias and A. R. Neto, Training soft margin support vector machines by simulated annealing: A dual approach, Expert Systems with Applications, 87 (2017), 157-169.
[11]	E. Erdogan and G. Iyengar, Ambiguous chance constrained problems and robust optimization, Math. Program., 107 (2006), 37-61. doi: 10.1007/s10107-005-0678-0.
[12]	P. M. Esfahani and D. Kuhn, Data-driven distributionally robust optimization using the Wasserstein metric: Performance guarantees and tractable reformulations, Math. Program., 171 (2018), 115-166. doi: 10.1007/s10107-017-1172-1.
[13]	R. Gao and A. J. Kleywegt, Distributionally robust stochastic optimization with wasserstein distance, preprint, arXiv: 1604.02199.
[14]	Y. Guo, K. Baker, E. Dall'Anese, Z. C. Hu and T. H. Summers, Data-based distributionally robust stochastic optimal power flow-Part I: Methodologies, IEEE Transactions on Power Systems, 34 (2019), 1483-1492. doi: 10.1109/TPWRS.2018.2878385.
[15]	Y. Guo, K. Baker, E. Dall'Anese, Z. C. Hu and T. H. Summers, Data-based distributionally robust stochastic optimal power flow-Part II: Case studies, IEEE Transactions on Power Systems, 34 (2018), 1493-1503.
[16]	P. Georg and W. David, Ambiguity in portfolio selection, Quant. Finance, 7 (2007), 435-442. doi: 10.1080/14697680701455410.
[17]	B. Han, C. Shang and D. Huang, Multiple kernel learning-aided robust optimization: Learning algorithm, computational tractability, and usage in multi-stage decision-making, European J. Oper. Res., 292 (2021), 1004-1018. doi: 10.1016/j.ejor.2020.11.027.
[18]	R. Jagannathan, Chance-constrained programming with joint constraints, Operations Res., 22 (1974), 358-372. doi: 10.1287/opre.22.2.358.
[19]	R. Ji and M. A. Lejeune, Data-driven optimization of reward-risk ratio measures, INFORMS J. Comput., 33 (2021), 1120-1137. doi: 10.1287/ijoc.2020.1002.
[20]	S. Justin and G. Leonidas, Convolutional wasserstein distances: Efficient optimal transportation on geometric domains, ACM Transactions on Graphics, 34 (2015), 1-11.
[21]	G. R. Lanckriet, L. Ghaoui, C. Bhattacharyya and M. I. Jordan, A robust minimax approach to classification, J. Mach. Learn. Res., 3 (2003), 555-582.
[22]	B. Piccoli and F. Rossi, On properties of the generalized Wasserstein distance, Arch. Ration. Mech. Anal., 222 (2016), 1339-1365. doi: 10.1007/s00205-016-1026-7.
[23]	P. K. Shivaswamy, C. Bhattacharyya and A. J. Smola, Second order cone programming approaches for handling missing and uncertain data, J. Mach. Learn. Res., 7 (2006), 1283-1314.
[24]	H. Xu, C. Caramanis and S. Mannor, Robustness and regularization of support vector machines, J. Mach. Learn. Res., 10 (2009), 1485-1510.
[25]	S. Zymler, D. Kuhn and B. Rustem, Distributionally robust joint chance constraints with second-order moment information, Math. Program., 137 (2013), 167-198. doi: 10.1007/s10107-011-0494-7.
[26]	S. Zymler, D. Kuhn and B. Rustem, Worst-case value at risk of nonlinear portfolios, Management Science, 59 (2013), 172-188. doi: 10.1287/mnsc.1120.1615.