• Journal of Intelligence and Information Systems
• /
• v.16 no.4
• /
• pp.99-112
• /
• 2010

Ensemble learning is a method for improving the performance of classification and prediction algorithms. It is a method for finding a highly accurateclassifier on the training set by constructing and combining an ensemble of weak classifiers, each of which needs only to be moderately accurate on the training set. Ensemble learning has received considerable attention from machine learning and artificial intelligence fields because of its remarkable performance improvement and flexible integration with the traditional learning algorithms such as decision tree (DT), neural networks (NN), and SVM, etc. In those researches, all of DT ensemble studies have demonstrated impressive improvements in the generalization behavior of DT, while NN and SVM ensemble studies have not shown remarkable performance as shown in DT ensembles. Recently, several works have reported that the performance of ensemble can be degraded where multiple classifiers of an ensemble are highly correlated with, and thereby result in multicollinearity problem, which leads to performance degradation of the ensemble. They have also proposed the differentiated learning strategies to cope with performance degradation problem. Hansen and Salamon (1990) insisted that it is necessary and sufficient for the performance enhancement of an ensemble that the ensemble should contain diverse classifiers. Breiman (1996) explored that ensemble learning can increase the performance of unstable learning algorithms, but does not show remarkable performance improvement on stable learning algorithms. Unstable learning algorithms such as decision tree learners are sensitive to the change of the training data, and thus small changes in the training data can yield large changes in the generated classifiers. Therefore, ensemble with unstable learning algorithms can guarantee some diversity among the classifiers. To the contrary, stable learning algorithms such as NN and SVM generate similar classifiers in spite of small changes of the training data, and thus the correlation among the resulting classifiers is very high. This high correlation results in multicollinearity problem, which leads to performance degradation of the ensemble. Kim,s work (2009) showedthe performance comparison in bankruptcy prediction on Korea firms using tradition prediction algorithms such as NN, DT, and SVM. It reports that stable learning algorithms such as NN and SVM have higher predictability than the unstable DT. Meanwhile, with respect to their ensemble learning, DT ensemble shows the more improved performance than NN and SVM ensemble. Further analysis with variance inflation factor (VIF) analysis empirically proves that performance degradation of ensemble is due to multicollinearity problem. It also proposes that optimization of ensemble is needed to cope with such a problem. This paper proposes a hybrid system for coverage optimization of NN ensemble (CO-NN) in order to improve the performance of NN ensemble. Coverage optimization is a technique of choosing a sub-ensemble from an original ensemble to guarantee the diversity of classifiers in coverage optimization process. CO-NN uses GA which has been widely used for various optimization problems to deal with the coverage optimization problem. The GA chromosomes for the coverage optimization are encoded into binary strings, each bit of which indicates individual classifier. The fitness function is defined as maximization of error reduction and a constraint of variance inflation factor (VIF), which is one of the generally used methods to measure multicollinearity, is added to insure the diversity of classifiers by removing high correlation among the classifiers. We use Microsoft Excel and the GAs software package called Evolver. Experiments on company failure prediction have shown that CO-NN is effectively applied in the stable performance enhancement of NNensembles through the choice of classifiers by considering the correlations of the ensemble. The classifiers which have the potential multicollinearity problem are removed by the coverage optimization process of CO-NN and thereby CO-NN has shown higher performance than a single NN classifier and NN ensemble at 1% significance level, and DT ensemble at 5% significance level. However, there remain further research issues. First, decision optimization process to find optimal combination function should be considered in further research. Secondly, various learning strategies to deal with data noise should be introduced in more advanced further researches in the future.

• Corrosion Science and Technology
• /
• v.8 no.2
• /
• pp.62-67
• /
• 2009

Ti and Ti-alloys have good biocompatibility, appropriate mechanical properties and excellent corrosion resistance. However, the widely used Ti-6Al-4V is found to release toxic ions (Al and V) into the body, leading to undesirable long-term effects. Ti-6Al-4V has much higher elastic modulus (100 GPa) than cortical bone (20 GPa). Therefore, titanium alloys with low elastic modulus have been developed as biomaterials to minimize stress shielding. The electrochemical behavior of surface-modified and MC3T3-E1 cell-cultured Ti-30(Nb,Ta) alloys with low elastic modulus have been investigated using various electrochemical methods. Surfaces of test samples were treated as follows: $0.3{\mu}m$ polished; $25{\mu}m$, $50{\mu}m$ and $125{\mu}m$ sandblasted. Specimen surfaces were cultured with MC3T3-E1 cells for 2 days. Average surface roughness ($R_a$) and morphology of specimens were determined using a surface profilometer, OM, and FE-SEM. Corrosion behavior was investigated using a potentiostat(EG&G PARSTAT 2273), and electrochemical impedance spectroscopy was performed (10 mHz to 100 kHz) in 0.9% NaCl solution at $36.5{\pm}1^{\circ}C$. The microstructures of the Ti-30(Ta,Nb) alloys had a needle-like appearance. The $R_a$ of polished Ti-30Ta and Ti-30Nb alloys was lower than that of the sandblasted Ti alloy. Cultured cells displayed round shapes. For polished alloy samples, cells were well-cultured on all surfaces compared to sandblasted alloy samples. In sandblasted and cell-cultured Ti-30(Nb,Ta) alloy, the pitting potential decreased and passive current density increased as $R_a$ increased. Anodic polarization curves of cell-cultured Ti alloys showed unstable behavior in the passive region compared to non-cell-cultured alloys. From impedance tests of sandblasted and cell-cultured alloys, the polarization resistance decreased as $R_a$ increased, whereas, $R_a$ for cell-cultured Ti alloys increased compared to non-cell-cultured Ti alloys.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.34 no.9
• /
• pp.859-866
• /
• 2010

Nonlinear dynamics of pulsating instability-diffusional-thermal instability with Lewis numbers sufficiently higher than unity-in counterflow diffusion flames, is numerically investigated by imposing a Damkohler number perturbation. The flame evolution exhibits three types of nonlinear behaviors, namely, decaying pulsating behavior, diverging behavior (which leads to extinction), and stable limit-cycle behavior. The stable limit-cycle behavior is observed in counterflow diffusion flames, but not in diffusion flames with a stagnant mixing layer. The critical value of the perturbed Damkohler number, which indicates the region where the three different flame behaviors can be observed, is obtained. A stable simple limit cycle, in which two supercritical Hopf bifurcations exist, is found in a narrow range of Damkohler numbers. As the flame temperature is increased, the stable simple limit cycle disappears and an unstable limit cycle corresponding to subcritical Hopf bifurcation appears. The period-doubling bifurcation is found to occur in a certain range of Damkohler numbers and temperatures, which leads to extend the lower boundary of supercritical Hopf bifurcation.

• Korean Chemical Engineering Research
• /
• v.47 no.1
• /
• pp.133-140
• /
• 2009

This paper presents the experimental study to analyze the pressure and production behavior using depressurization and thermal methods in order to evaluate the hydrate productivity in the 2-D multi-well, plate-type apparatus which has 80 md permeability and 30% hydrate saturation. Injecting methane gas through multi-well allowed to set up the highly saturated hydrate system and combining two different sorts of sands made possible to build up the low permeability system. In this system, both depressurization and electric stimulation methods were applied. When operating pressure was low, according to the depressurization experiments results, the gas recovery was high, however strong pulses which appeared at initial stage of production would damage the operation system. Moreover, cases that hydrate reformed have occurred by endothermic reaction. We have conducted experiments four and six times for the depressurization magnitudes of 140 psi and 320 psi, respectively, to analyze production behavior for the method more in detail. For the cases that the depressurization magnitude was set as 140 psi, the unstable period appears in the results, but stabilized soon. In the experiment results for 320 psi the discontinuous and intermittent behavior has been observed. Thermal stimulation experiments was conducted with depressurizing 80 psi which is the case that shows stable behavior and low recovery. In the results, the gas recovery was high and the energy efficiency was low for long stimulating time. The energy efficiency and gas recovery increased for the soaking time of 1 minute after 2 minute-preheating. In the cases of which the soaking time exceeds 1 minute, energy loss by long soaking time caused low gas recovery and poor energy efficiency.

• Journal of the Korea Concrete Institute
• /
• v.13 no.6
• /
• pp.568-574
• /
• 2001

From measured responses of concrete three-point bend tests, the average values of the responses have been calculated. The fracture behavior of continuously propagating concrete crack has been analyzed from the average responses. The experimental parameters of this study were the initial notch sizes of 25.4㎜ and 6.4 ㎜ and the processing times of 2,000 sec. and 20 sec . The different notch sizes were used for the effects of the size of fracture process zone and specimen geometry, and the processing times for those of initial creep. However the load-point displacement rate in this study did not affect the experimental responses seriously. The average loads were calculated from the average external work of a series of tests, and average crack lengths were determined by using strain gages. Before the peak load, the resistance curve could be determined from the size of fracture process zone, but unstable crack propagation of 88㎜ occurred at the load-point displacement of 0.088∼0.154㎜ after the peak load. The average fracture energy density G$\_$F/$\^$ave/ = 115 N/m occurred during the unstable crack propagation. The fracture process zones were fully developed at the crack length of 111㎜, and the sizes of fracture process zone for initial notches of 25.4㎜ and 6.4㎜ were 86㎜ and 105㎜, respectively. Average fracture energy densities of the resistance curves after full development of fracture process zone were 229 N/m for the initial notch of 25.4㎜ and 284 N/m for 6.4㎜. The values were more than twice of G$\_$F/$\^$ave/.

• Journal of the Korea Concrete Institute
• /
• v.14 no.5
• /
• pp.689-697
• /
• 2002

The dynamic loads and load-point displacements of concrete three-point bend (TPB) specimens had been measured. The average crack velocities measured with strain gages were 0.16 ㎜/sec ∼ 66 m/sec. The fracture energy for crack extension was determined from the difference of the kinetic energy for the load-point velocity and the strain energy without permanent deformation from the measure external work. For all crack velocities, there were micro-cracking for 23 ㎜ crack extension, stable cracking for 61 ㎜ crack extension at the maximum strain energy, and then unstable cracking. The unstable crack extension was arrested at 80 ㎜ crack extension except the tests of 66 m/sec crack velocity. The tests less than 13 ㎜/sec crack velocity and faster than 1.9 m/sec showed static and dynamic fracture behaviors, respectively. In spite of much difference of the load and load-point displacement relations for the crack velocities, the crack velocities of dynamic tests did not affect on fracture energy rate during the stable crack extension due to the reciprocal action of kinetic force, crack extension and strain energy. During stable crack extension, the maximum fracture resistances of the dynamic tests was 147％ larger than that of the static tests.

• Proceedings of the KWS Conference
• /
• 2002.10a
• /
• pp.612-619
• /
• 2002

The dynamic behavior of Al-Mg alloys plasma was very unstable and this instability was closely related to the unstable motion of keyhole during laser irradiation. The keyhole fluctuated both in size and shape and its fluctuation period was about 440 ${\mu}{\textrm}{m}$. This instability has been estimated to be caused by the evaporation phenomena of metals with different boiling point and latent heats of vaporization. Therefore, the authors have conducted the spectroscopic diagnostics of plasma induced in the pulsed YAG laser welding of Al-Mg alloys in air and argon atmospheres. In the air environment, the identified spectra were atomic lines of Al, Mg, Cr, Mn, Cu, Fe and Zn, and singly ionized Mg line, as well as strong molecular spectrum of AlO, MgO and AIH. It was confirmed that the resonant lines of Al and Mg were strongly self-absorbed, in particular in the vicinity of pool surface. The self-absorption of atomic Mg line was more eminent in alloys containing higher Mg. These facts showed that the laser-induced plasma was relatively a low temperature and high density metallic vapor. The intensities of molecular spectra of AlO and MgO were different each other depending on the power density of laser beam. Under the low power density irradiation condition, the MgO band spectra were predominant in intensity, while the AlO spectra became much stronger in higher power density. In argon atmosphere the band spectra of MgO and AlO completely vanished, but AlH molecular spectra was detected clearly. The hydrogen source was presumably the hydrogen solved in the base Metal, absorbed water on the surface oxide layer or H$_2$ and $H_2O$ in the shielding gas. The temporal change in spectral line intensities was quite similar to the fluctuation of keyhole. The time average plasma temperature at 1 mm high above the surface of A5083 alloy was determined by the Boltzmann plot method of atomic Cr lines of different excitation energy. The obtained electron temperature was 3, 280$\pm$150 K which was about 500 K higher than the boiling point of pure aluminum. The electron number density was determined by measuring the relative intensities of the spectra1lines of atomic and singly ionized Magnesium, and the obtained value was 1.85 x 1019 1/㎥.

• Journal of Korean Society of Steel Construction
• /
• v.24 no.3
• /
• pp.337-348
• /
• 2012

This study investigated the characteristics of bifurcation and the instability due to the initial imperfection of the space truss, which is sensitive to the initial conditions, and the calculated buckling load by the analysis of Eigen-values and the determinant of tangential stiffness. A two-free nodes model, a star dome, and a three-ring dome model were selected as case studies in order to examine the unstable phenomenon due to the sensitivity to Eigen mode, and the influence of the rise-span ratio and the load parameter on the buckling load were analyzed. The sensitivity to the imperfection of the two-free nodes model changed the critical path after reaching the limit point through the bifurcation mode, and the buckling load level was reduced by the increase in the amount of imperfection. The two sensitive buckling patterns for the model can be explained by investigating the displaced position of the free node, and the asymmetric Eigen mode was a major influence on the unstable behavior due to the initial imperfection. The sensitive mode was similar to the in-extensional mechanism basis of the simplified model. Since the rise-span ratio was higher, the effect of local buckling is more prominent than the global buckling in the star dome, and bifurcation on the equilibrium path occurring as the value of the load parameter was higher. Additionally, the buckling load levels of the star dome and the three-ring model were about 50-70% and 80-90% of the limit point, respectively.

• Journal of the Korean Society of Food Culture
• /
• v.16 no.2
• /
• pp.147-157
• /
• 2001

In order to investigate the eating out behavior of university students, this survey was conducted using the questionaires for 710 students(369 male, 341 female) from 11 universities in Seoul. It was revealed that 39.4% of the subjects spent $60,000{\sim}100,000$ won for monthly eating out cost and 57.8% of them ate out more than once a day. Most of them expended less than 3,500 won for lunch, while 36.5% of them spent $4,000 {\sim}5,000$ won for dinner. Dinner was regarded more important than lunch. Korean foods were the most preferred menu for eating out with friends and fast foods were the second. But Boonsik(snack bar foods), Chinese foods and Japanease foods were rarely selected. Frequency of selecting fast foods was 8 times greater than that of Boonsik. This indicates that the preference of western flavor and the pursuit of convenience is getting more obvious. While dating, western foods were preferred, followed by Korean foods, fast foods. The 80 kinds of foods were reported as favored eating out foods. Although 50 among 80 were Korean foods, the rest of them were Koreanized foreign foods most of those were western style. This may suggest that when the students become adults, they will be much fond of western dish for their dinning out. This tendency of preferring western flavor were much apparent in foods for dinner compared with lunch. In both sexes, the standard of food choice was in the order of taste, price, mood, hygiene, service and brand name. But male students were more conscious of price and service while female students were more concerned about taste and hygiene. Most unsatisfying feature in restaurant was unstable atmosphere for both sexes. Taste was the most important sensory factor in selecting the foods, followed by appearance, smell and texture. Major source of restaurant information was recommendation by friends or relatives. But the use of internet or magazine was negligible. Female students had more positive attitude, compared with male students, in using restaurant information and pursuing eating out for gourmet. The dining out menu of which price ranges about $3,000{\sim}5,000$ won could be preferable foods for most people. Therefore, instead of blaming them for eating too much fast foods, new menus which fit the food preference and affordability of the students should be developed.

• Journal of Nutrition and Health
• /
• v.42 no.4
• /
• pp.358-365
• /
• 2009

The study was performed for investigation with 257 female adults to find out the factors affecting health control behavior and obesity stress by personality type. First of all, BMIs of the respondents showed 30 low weight (11.7%), 170 normal weight (66.1%), 33 overweight (12.8%) and 24 obese respondents (9.4%). Comparing with low weight or normal weight, overweight or obesity could be seen as strong with diet control behaviors, from which obesity stress showed the tendency to be high accordingly. Second, the result of the investigation showed that for the open personality, which was featured as strongly desired to accomplish, very sincere and sensitive, they were more eager to control through health or diet. Third, the result of the investigation showed that for the more nervous, which is featured as emotionally unstable, bad at self control of impetus and clumsy to handle stress, the obesity stress was escalating.