Advanced Search
Article Contents
Article Contents

# Two penalized mixed–integer nonlinear programming approaches to tackle multicollinearity and outliers effects in linear regression models

• * Corresponding author: Mahdi Roozbeh
• In classical regression analysis, the ordinary least–squares estimation is the best strategy when the essential assumptions such as normality and independency to the error terms as well as ignorable multicollinearity in the covariates are met. However, if one of these assumptions is violated, then the results may be misleading. Especially, outliers violate the assumption of normally distributed residuals in the least–squares regression. In this situation, robust estimators are widely used because of their lack of sensitivity to outlying data points. Multicollinearity is another common problem in multiple regression models with inappropriate effects on the least–squares estimators. So, it is of great importance to use the estimation methods provided to tackle the mentioned problems. As known, robust regressions are among the popular methods for analyzing the data that are contaminated with outliers. In this guideline, here we suggest two mixed–integer nonlinear optimization models which their solutions can be considered as appropriate estimators when the outliers and multicollinearity simultaneously appear in the data set. Capable to be effectively solved by metaheuristic algorithms, the models are designed based on penalization schemes with the ability of down–weighting or ignoring unusual data and multicollinearity effects. We establish that our models are computationally advantageous in the perspective of the flop count. We also deal with a robust ridge methodology. Finally, three real data sets are analyzed to examine performance of the proposed methods.

Mathematics Subject Classification: Primary: 62J05, 90C11; Secondary: 90C59.

 Citation:

• Figure 1.  The diagnostic plots of the model (18)

Figure 2.  The diagram of ${\rm GCV}(k,z)$ versus the ridge parameter for the bridge projects data set

Figure 3.  The diagnostic plots for the model (20)

Figure 4.  The diagram of ${\rm GCV}(k,z)$ versus the ridge parameter for the electricity data

Figure 5.  The diagnostic plots for the model (21)

Figure 6.  The diagram of ${\rm GCV}(k,z)$ versus the ridge parameter for the CPS data

Table 1.  Evaluation of the proposed estimators for the bridge projects data set

 Method Coefficients OLS RLTS MLTSCM UBDMLTSCM1 $Intercept$ 2.3317 1.91363 2.0304 1.8278 $\log(CCost)$ 0.1483 0.33718 0.3056 0.2923 $\log(Dwgs)$ 0.8356 0.58002 0.6210 0.7829 $\log(Spans)$ 0.1963 0.06662 0.0657 0.0241 ${\rm SSE}$ 3.8692 1.9788 1.9778 1.0577 ${\rm R}^2$ 0.7747 0.8579 0.8600 0.9147 Method Coefficients UBDMLTSCM2 LSVR NSVR NNR $Intercept$ 1.9140 -0.0125 - -7.8431 $\log(CCost)$ 0.2360 0.4152 - 0.4236 $\log(Dwgs)$ 0.8914 0.3933 - 2.8061 $\log(Spans)$ 0.0467 0.1176 - 0.5110 ${\rm SSE}$ 1.1504 4.0131 2.7834 1.7108 ${\rm R}^2$ 0.9020 0.7663 0.8379 0.9004

Table 2.  The most effective subgroup of predictor variables based on the ${\rm R}^2_{adj}$ and AIC criteria for the electricity data set

 Subset size Predictor variables ${\rm R}^2_{adj}$ AIC 1 $Temp$ 0.5523 -1067.814 2 $Temp,LREG$ 0.5781 -1077.339 3 ${\bf Temp,LREG,LI}$ 0.5892 -1081.063 4 $Temp,LREG,LI,x_{9}$ 0.5891 -1080.057 5 $Temp,LREG,LI,x_{9},x_{10}$ 0.5882 -1078.709 6 $Temp,LREG,LI,x_{9},x_{10},x_{11}$ 0.5875 -1077.427 7 $Temp,LREG,LI,x_{9},x_{10},x_{11},x_{1}$ 0.5858 -1075.734 8 $Temp,LREG,LI,x_{9},x_{10},x_{11},x_{1},x_{3}$ 0.5837 -1073.897 9 $Temp,LREG,LI,x_{9},x_{10},x_{11},x_{1},x_{3},x_{5}$ 0.5812 -1071.907 10 $Temp,LREG,LI,x_{9},x_{10},x_{11},x_{1},x_{3},x_{5},x_{4}$ 0.5789 -1069.987 11 $Temp,LREG,LI,x_{9},x_{10},x_{11},x_{1},x_{3},x_{5},x_{4},x_{7}$ 0.5764 -1067.997 12 $Temp,LREG,LI,x_{9},x_{10},x_{11},x_{1},x_{3},x_{5},x_{4},x_{7},x_{2}$ 0.5740 -1064.098 13 $Temp,LREG,LI,x_{9},x_{10},x_{11},x_{1},x_{3},x_{5},x_{4},x_{7},x_{2},x_{6}$ 0.5718 -1064.281 14 $Temp,LREG,LI,x_{9},x_{10},x_{11},x_{1},x_{3},x_{5},x_{4},x_{7},x_{2},x_{6},x_{8}$ 0.5709 -1063.014

Table 3.  Evaluation of the proposed estimators for the electricity data set

 Method Coefficients OLS RLTS MLTSCM UBDMLTSCM1 $Intercept$ 4.4069 5.1693 4.9881 5.2039 $LI$ 0.1925 0.0989 0.1146 0.0956 $LREG$ -0.0778 -0.0939 -0.1054 -0.0956 $Temp$ -0.0002 -0.0002 -0.0003 -0.0003 ${\rm SSE}$ 0.3765 0.2637 0.1982 0.1296 ${\rm R}^2$ 0.5962 0.6742 0.7399 0.7559 Method Coefficients UBDMLTSCM2 LSVR NSVR NNR $Intercept$ 4.0907 0.0881 - 2.6215 $LI$ 0.2225 0.1545 - 1.2806 $LREG$ -0.0940 -0.1322 - -3.7418 $Temp$ -0.0003 -0.7508 - -0.8067 ${\rm SSE}$ 0.1413 0.3881 0.2629 0.4240 ${\rm R}^2$ 0.7468 0.5838 0.7181 0.5452

Table 4.  Evaluation of the proposed estimators for the CPS data

 Method Coefficients OLS RLTS MLTSCM UBDMLTSCM1 $Intercept$ 1.0786 0.7498 1.1963 0.9257 $education$ 0.1794 0.1482 0.2576 0.2018 $south$ -0.1024 -0.1208 -0.1109 -0.1174 $sex$ -0.2220 -0.2851 -0.2776 -0.2665 $experience$ 0.0958 0.0613 0.1630 0.1090 $union$ 0.2005 0.1939 0.1987 0.1427 $age$ -0.0854 -0.0473 -0.1510 -0.0960 $race$ 0.0504 0.0674 0.0482 0.0749 $occupation$ -0.0074 -0.0122 0.0072 -0.0126 $sector$ 0.0915 0.0614 0.0411 0.0965 $married$ 0.0766 0.0590 0.1937 0.0924 ${\rm SSE}$ 101.17 76.3827 50.5810 49.8101 ${\rm R}^2$ 0.3185 0.4049 0.4146 0.4123 Method Coefficients UBDMLTSCM2 LSVR NSVR NNR $Intercept$ 0.9038 0.0054 - -5.5913 $education$ 0.1974 0.4997 - 0.6978 $south$ -0.0916 -0.1141 - -0.4331 $sex$ -0.2416 -0.2638 - -0.9731 $experience$ 0.1011 0.2573 - 0.2991 $union$ 0.1791 0.1511 - 1.0483 $age$ -0.0888 0.0420 - -0.2590 $race$ 0.0515 0.0930 - 0.2437 $occupation$ -0.0140 -0.0526 - 0.0004 $sector$ 0.0810 0.0918 - 0.3258 $married$ 0.1216 0.0524 - 0.4156 ${\rm SSE}$ 49.2827 102.5847 79.0911 84.2234 ${\rm R}^2$ 0.4279 0.3089 0.4672 0.4326
•  [1] E. H. L. Aarts, J. H. M. Korst and P. J. M. van Laarhoren, Simulated annealing, in Local Search in Combinatorial Optimization, Wiley-Intersci. Ser. Discrete Math. Optim., Wiley-Intersci. Publ., Wiley, Chichester, 1997, 91–121. [2] E. Akdenïz Duran, W. K. Härdle and M. Osipenko, Difference based ridge and Liu type estimators in semiparametric regression models, J. Multivariate Anal., 105 (2012), 164-175.  doi: 10.1016/j.jmva.2011.08.018. [3] F. Akdenïz and M. Roozbeh, Generalized difference-based weighted mixed almost unbiased ridge estimator in partially linear models, Statist. Papers, 60 (2019), 1717-1739.  doi: 10.1007/s00362-017-0893-9. [4] M. Amini and M. Roozbeh, Optimal partial ridge estimation in restricted semiparametric regression models, J. Multivariate Anal., 136 (2015), 26-40.  doi: 10.1016/j.jmva.2015.01.005. [5] M. Arashi and T. Valizadeh, Performance of Kibria's methods in partial linear ridge regression model, Statist. Pap., 56 (2015), 231-246.  doi: 10.1007/s00362-014-0578-6. [6] M. Awad and R. Khanna, Efficient Learning Machines: Theories, Concepts, and Applications for Engineers and System Designers, Apress, Berkeley, CA, 2015. doi: 10.1007/978-1-4302-5990-9. [7] S. Babaie–Kafaki, R. Ghanbari and N. Mahdavi–Amiri, An efficient and practically robust hybrid metaheuristic algorithm for solving fuzzy bus terminal location problems, Asia–Pac. J. Oper. Res., 29 (2012), 1-25.  doi: 10.1142/S0217595912500091. [8] S. Babaie-Kafaki, R. Ghanbari and N. Mahdavi-Amiri, Hybridizations of genetic algorithms and neighborhood search metaheuristics for fuzzy bus terminal location problems, Appl. Soft Comput., 46 (2016), 220-229.  doi: 10.1016/j.asoc.2016.03.005. [9] S. Roozbeh and M. Babaie-Kafakiand, A revised Cholesky decomposition to combat multicollinearity in multiple regression models, J. Stat. Comput. Simul., 87 (2017), 2298-2308.  doi: 10.1080/00949655.2017.1328599. [10] M. R. Baye and D. F. Parker, Combining ridge and principal component regression: A money demand illustration, Comm. Statist. A—Theory Methods, 13 (1984), 197-205.  doi: 10.1080/03610928408828675. [11] E. R. Berndt, The Practice of Econometrics, New York, Addison-Wesley, 1991. [12] D. Bertsimas and J. N. Tsitsiklis, Introduction to Linear Optimization, Athena Scientific, Massachusetts, 1997. [13] P. Bühlmann, M. Kalisch and L. Meier, High–dimensional statistics with a view towards applications in biology, Ann. Rev. Stat. Appl., 1 (2014), 255-278. [14] R. H. Byrd and J. Nocedal, A tool for the analysis of quasi–Newton methods with application to unconstrained minimization, SIAM J. Numer. Anal., 26 (1989), 727-739.  doi: 10.1137/0726042. [15] M. Hassanzadeh Bashtian, M. Arashi and S. M. M. Tabatabaey, Using improved estimation strategies to combat multicollinearity, J. Stat. Comput. Simul., 81 (2011), 1773-1797.  doi: 10.1080/00949655.2010.505925. [16] S. Hawkins, H. He, G. Williams and R. Baxter, Outlier detection using replicator neural networks, in International Conference on Data Warehousing and Knowledge Discovery, Springer, Berlin, Heidelberg, (2002), 170–180. doi: 10.1007/3-540-46145-0_17. [17] D. Henderson, S. H. Jacobson and A. W. Johnson, The theory and practice of simulated annealing, in Handbook of Metaheuristics, Kluwer Academic Publishers, Boston, MA, (2003), 287–319. doi: 10.1007/0-306-48056-5_10. [18] A. E. Hoerl and R. W. Kennard, Ridge regression: Biased estimation for non–orthogonal problems, Technometrics, 12 (1970), 55-67. [19] P. W. Holland and R. E. Welsch, Robust regression using iteratively reweighted least–squares, Comm. Statist. Theo. Meth., 6 (1977), 813-827. [20] G. James, D. Witten, T. Hastie and R. Tibshirani, An Introduction to Statistical Learning, Springer, New York, 2013. doi: 10.1007/978-1-4614-7138-7. [21] S. Kaçiranlar and S. Sakallioǧlu, Combining the Liu estimator and the principal component regression estimator, Comm. Statist. Theory Methods, 30 (2001), 2699-2705.  doi: 10.1081/STA-100108454. [22] A. Karatzoglou, D. Meyer and K. Hornik, Support Vector Machines in R, J. Stat. Softw., 15 (2006), 1-28. [23] K. J. Liu, A new class of biased estimate in linear regression, Comm. Statist. Theory Methods, 22 (1993), 393-402.  doi: 10.1080/03610929308831027. [24] A. Mohammad Nezhad, R. Aliakbari Shandiz and A. H. Eshraghniaye Jahromi, A particle swarm–BFGS algorithm for nonlinear programming problems, Comput. Oper. Res., 40 (2013), 963-972.  doi: 10.1016/j.cor.2012.11.008. [25] G. Piazza and T. Politi, An upper bound for the condition number of a matrix in spectral norm, J. Comput. Appl. Math., 143 (2002), 141-144.  doi: 10.1016/S0377-0427(02)00396-5. [26] W. M. Pride and O. C. Ferrel, Marketing, 15th edition, South-Western, Cengage Learning, International Edition, 2010. [27] C. R. Reeves, Modern heuristic techniques, in Modern Heuristic Search Methods, John Wiley and Sons, Chichester, (1996), 1–24. [28] M. Roozbeh, Optimal QR-based estimation in partially linear regression models with correlated errors using GCV criterion, Computational Statistics & Data Analysis, 117 (2018), 45-61.  doi: 10.1016/j.csda.2017.08.002. [29] M. Roozbeh, S. Babaie-Kafaki and M. Arashi, A class of biased estimators based on QR decomposition, Linear Algebra Appl., 508 (2016), 190-205.  doi: 10.1016/j.laa.2016.07.009. [30] M. Roozbeh, S. Babaie-Kafaki and A. Naeimi Sadigh, A heuristic approach to combat multicollinearity in least trimmed squares regression analysis, Appl. Math. Model, 57 (2018), 105-120.  doi: 10.1016/j.apm.2017.11.011. [31] M. Roozbeh, Robust ridge estimator in restricted semiparametric regression models, J. Multivariate Anal., 147 (2016), 127-144.  doi: 10.1016/j.jmva.2016.01.005. [32] P. J. Rousseeuw, Least median of squares regression, J. Amer. Statist. Assoc., 79 (1984), 871-880.  doi: 10.1080/01621459.1984.10477105. [33] P. J. Rousseeuw, and A. M. Leroy, Robust Regression and Outlier Detection, John Wiley and Sons, New York, 1987. doi: 10.1002/0471725382. [34] S. J. Sheather, A Modern Approach to Regression with R, Springer, New York, 2009. doi: 10.1007/978-0-387-09608-7. [35] W. Sun and Y. X. Yuan, Optimization Theory and Methods: Nonlinear Programming, Springer, New York, 2006. [36] P. Tryfos, Methods for Business Analysis and Forecasting: Text & Cases, John Wiley and Sons, New York, 1998. [37] D. S. Watkins, Fundamentals of Matrix Computations, 2nd edition, John Wiley and Sons, New York, 2002. doi: 10.1002/0471249718. [38] X. S. Yang, Nature–Inspired Optimization Algorithms, Elsevier, Amsterdam, 2014.

Figures(6)

Tables(4)

## Article Metrics

HTML views(1140) PDF downloads(386) Cited by(0)

## Other Articles By Authors

• on this site
• on Google Scholar

### Catalog

/

DownLoad:  Full-Size Img  PowerPoint