• Journal of Intelligence and Information Systems
• /
• v.18 no.4
• /
• pp.19-42
• /
• 2012

Consumer consumption patterns are shifting rapidly as buyers migrate from offline markets to e-commerce routes, such as shopping channels on TV and internet shopping malls. In the offline markets consumers go shopping, see the shopping items, and choose from them. Recently consumers tend towards buying at shopping sites free from time and place. However, as e-commerce markets continue to expand, customers are complaining that it is becoming a bigger hassle to shop online. In the online shopping, shoppers have very limited information on the products. The delivered products can be different from what they have wanted. This case results to purchase cancellation. Because these things happen frequently, they are likely to refer to the consumer reviews and companies should be concerned about consumer's voice. E-commerce is a very important marketing tool for suppliers. It can recommend products to customers and connect them directly with suppliers with just a click of a button. The recommender system is being studied in various ways. Some of the more prominent ones include recommendation based on best-seller and demographics, contents filtering, and collaborative filtering. However, these systems all share two weaknesses : they cannot recommend products to consumers on a personal level, and they cannot recommend products to new consumers with no buying history. To fix these problems, we can use the information which has been collected from the questionnaires about their demographics and preference ratings. But, consumers feel these questionnaires are a burden and are unlikely to provide correct information. This study investigates combining collaborative filtering with the centrality of social network analysis. This centrality measure provides the information to infer the preference of new consumers from the shopping history of existing and previous ones. While the past researches had focused on the existing consumers with similar shopping patterns, this study tried to improve the accuracy of recommendation with all shopping information, which included not only similar shopping patterns but also dissimilar ones. Data used in this study, Movie Lens' data, was made by Group Lens research Project Team at University of Minnesota to recommend movies with a collaborative filtering technique. This data was built from the questionnaires of 943 respondents which gave the information on the preference ratings on 1,684 movies. Total data of 100,000 was organized by time, with initial data of 50,000 being existing customers and the latter 50,000 being new customers. The proposed recommender system consists of three systems : [+] group recommender system, [-] group recommender system, and integrated recommender system. [+] group recommender system looks at customers with similar buying patterns as 'neighbors', whereas [-] group recommender system looks at customers with opposite buying patterns as 'contraries'. Integrated recommender system uses both of the aforementioned recommender systems to recommend movies that both recommender systems pick. The study of three systems allows us to find the most suitable recommender system that will optimize accuracy and customer satisfaction. Our analysis showed that integrated recommender system is the best solution among the three systems studied, followed by [-] group recommended system and [+] group recommender system. This result conforms to the intuition that the accuracy of recommendation can be improved using all the relevant information. We provided contour maps and graphs to easily compare the accuracy of each recommender system. Although we saw improvement on accuracy with the integrated recommender system, we must remember that this research is based on static data with no live customers. In other words, consumers did not see the movies actually recommended from the system. Also, this recommendation system may not work well with products other than movies. Thus, it is important to note that recommendation systems need particular calibration for specific product/customer types.

• Journal of Internet Computing and Services
• /
• v.14 no.5
• /
• pp.69-76
• /
• 2013

The incidence of globally infectious and pathogenic diseases such as H1N1 (swine flu) and Avian Influenza (AI) has recently increased. An infectious disease is a pathogen-caused disease, which can be passed from the infected person to the susceptible host. Pathogens of infectious diseases, which are bacillus, spirochaeta, rickettsia, virus, fungus, and parasite, etc., cause various symptoms such as respiratory disease, gastrointestinal disease, liver disease, and acute febrile illness. They can be spread through various means such as food, water, insect, breathing and contact with other persons. Recently, most countries around the world use a mathematical model to predict and prepare for the spread of infectious diseases. In a modern society, however, infectious diseases are spread in a fast and complicated manner because of rapid development of transportation (both ground and underground). Therefore, we do not have enough time to predict the fast spreading and complicated infectious diseases. Therefore, new system, which can prevent the spread of infectious diseases by predicting its pathway, needs to be developed. In this study, to solve this kind of problem, an integrated monitoring system, which can track and predict the pathway of infectious diseases for its realtime monitoring and control, is developed. This system is implemented based on the conventional mathematical model called by 'Susceptible-Infectious-Recovered (SIR) Model.' The proposed model has characteristics that both inter- and intra-city modes of transportation to express interpersonal contact (i.e., migration flow) are considered. They include the means of transportation such as bus, train, car and airplane. Also, modified real data according to the geographical characteristics of Korea are employed to reflect realistic circumstances of possible disease spreading in Korea. We can predict where and when vaccination needs to be performed by parameters control in this model. The simulation includes several assumptions and scenarios. Using the data of Statistics Korea, five major cities, which are assumed to have the most population migration have been chosen; Seoul, Incheon (Incheon International Airport), Gangneung, Pyeongchang and Wonju. It was assumed that the cities were connected in one network, and infectious disease was spread through denoted transportation methods only. In terms of traffic volume, daily traffic volume was obtained from Korean Statistical Information Service (KOSIS). In addition, the population of each city was acquired from Statistics Korea. Moreover, data on H1N1 (swine flu) were provided by Korea Centers for Disease Control and Prevention, and air transport statistics were obtained from Aeronautical Information Portal System. As mentioned above, daily traffic volume, population statistics, H1N1 (swine flu) and air transport statistics data have been adjusted in consideration of the current conditions in Korea and several realistic assumptions and scenarios. Three scenarios (occurrence of H1N1 in Incheon International Airport, not-vaccinated in all cities and vaccinated in Seoul and Pyeongchang respectively) were simulated, and the number of days taken for the number of the infected to reach its peak and proportion of Infectious (I) were compared. According to the simulation, the number of days was the fastest in Seoul with 37 days and the slowest in Pyeongchang with 43 days when vaccination was not considered. In terms of the proportion of I, Seoul was the highest while Pyeongchang was the lowest. When they were vaccinated in Seoul, the number of days taken for the number of the infected to reach at its peak was the fastest in Seoul with 37 days and the slowest in Pyeongchang with 43 days. In terms of the proportion of I, Gangneung was the highest while Pyeongchang was the lowest. When they were vaccinated in Pyeongchang, the number of days was the fastest in Seoul with 37 days and the slowest in Pyeongchang with 43 days. In terms of the proportion of I, Gangneung was the highest while Pyeongchang was the lowest. Based on the results above, it has been confirmed that H1N1, upon the first occurrence, is proportionally spread by the traffic volume in each city. Because the infection pathway is different by the traffic volume in each city, therefore, it is possible to come up with a preventive measurement against infectious disease by tracking and predicting its pathway through the analysis of traffic volume.

• Journal of Astronomy and Space Sciences
• /
• v.22 no.2
• /
• pp.147-174
• /
• 2005

To understand the physical processes that control the high-latitude lower thermospheric dynamics, we quantify the forces that are mainly responsible for maintaining the high-latitude lower thermospheric wind system with the aid of the National Center for Atmospheric Research Thermosphere-Ionosphere Electrodynamics General Circulation Model (NCAR-TIEGCM). Momentum forcing is statistically analyzed in magnetic coordinates, and its behavior with respect to the magnitude and orientation of the interplanetary magnetic field (IMF) is further examined. By subtracting the values with zero IMF from those with non-zero IMF, we obtained the difference winds and forces in the high-latitude 1ower thermosphere(<180 km). They show a simple structure over the polar cap and auroral regions for positive($B_y$ > 0.8|$\overline{B}_z$ |) or negative($B_y$ < -0.8|$\overline{B}_z$|) IMF-$\overline{B}_y$ conditions, with maximum values appearing around -80$^{\circ}$ magnetic latitude. Difference winds and difference forces for negative and positive $\overline{B}_y$ have an opposite sign and similar strength each other. For positive($B_z$ > 0.3125|$\overline{B}_y$|) or negative($B_z$ < -0.3125|$\overline{B}_y$|) IMF-$\overline{B}_z$ conditions the difference winds and difference forces are noted to subauroral latitudes. Difference winds and difference forces for negative $\overline{B}_z$ have an opposite sign to positive $\overline{B}_z$ condition. Those for negative $\overline{B}_z$ are stronger than those for positive indicating that negative $\overline{B}_z$ has a stronger effect on the winds and momentum forces than does positive $\overline{B}_z$ At higher altitudes(>125 km) the primary forces that determine the variations of tile neutral winds are the pressure gradient, Coriolis and rotational Pedersen ion drag forces; however, at various locations and times significant contributions can be made by the horizontal advection force. On the other hand, at lower altitudes(108-125 km) the pressure gradient, Coriolis and non-rotational Hall ion drag forces determine the variations of the neutral winds. At lower altitudes(<108 km) it tends to generate a geostrophic motion with the balance between the pressure gradient and Coriolis forces. The northward component of IMF By-dependent average momentum forces act more significantly on the neutral motion except for the ion drag. At lower altitudes(108-425 km) for negative IMF-$\overline{B}_y$ condition the ion drag force tends to generate a warm clockwise circulation with downward vertical motion associated with the adiabatic compress heating in the polar cap region. For positive IMF-$\overline{B}_y$ condition it tends to generate a cold anticlockwise circulation with upward vertical motion associated with the adiabatic expansion cooling in the polar cap region. For negative IMF-$\overline{B}_z$ the ion drag force tends to generate a cold anticlockwise circulation with upward vertical motion in the dawn sector. For positive IMF-$\overline{B}_z$ it tends to generate a warm clockwise circulation with downward vertical motion in the dawn sector.

• Progress in Medical Physics
• /
• v.23 no.1
• /
• pp.15-25
• /
• 2012

The aim of this study is to evaluate plan quality and dose accuracy for Volumetric Modulated Arc Therapy (VMAT) on the TG-119 and is to investigate the effects on variation of the selectable optimization parameters of VMAT. VMAT treatment planning was implemented on a Varian iX linear accelerator with ARIA record and verify system (Varian Mecical System Palo Alto, CA) and Oncentra MasterPlan treatment planning system (Nucletron BV, Veenendaal, Netherlands). Plan quality and dosimetric accuracy were evaluated by effect of varying a number of arc, gantry spacing and delivery time for the test geometries provided in TG-119. Plan quality for the target and OAR was evaluated by the mean value and the standard deviation of the Dose Volume Histograms (DVHs). The ionization chamber and $Delta^{4PT}$ bi-planar diode array were used for the dose evaluation. For treatment planning evaluation, all structure sets closed to the goals in the case of single arc, except for the C-shape (hard), and all structure sets achieved the goals in the case of dual arc, except for C-shape (hard). For the variation of a number of arc, the simple structure such as a prostate did not have the difference between single arc and dual arc, whereas the complex structure such as a head and neck showed a superior result in the case of dual arc. The dose distribution with gantry spacing of $4^{\circ}$ was shown better plan quality than the gantry spacing of $6^{\circ}$, but was similar results compared with gantry spacing of $2^{\circ}$. For the verification of dose accuracy with single arc and dual arc, the mean value of a relative error between measured and calculated value were within 3% and 4% for point dose and confidence limit values, respectively. For the verification on dose accuracy with the gantry intervals of $2^{\circ}$, $4^{\circ}$ and $6^{\circ}$, the mean values of relative error were within 3% and 5% for point dose and confidence limit values, respectively. In the verification of dose distribution with $Delta^{4PT}$ bi-planar diode array, gamma passing rate was $98.72{\pm}1.52%$ and $98.3{\pm}1.5%$ for single arc and dual arc, respectively. The confidence limit values were within 4%. The smaller the gantry spacing, the more accuracy results were shown. In this study, we performed the VMAT QA based on TG-119 procedure, and demonstrated that all structure sets were satisfied with acceptance criteria. And also, the results for the selective optimization variables informed the importance of selection for the suitable variables according to the clinical cases.

• Tuberculosis and Respiratory Diseases
• /
• v.49 no.1
• /
• pp.37-48
• /
• 2000

Backgrounds : The pathophysiology of chronic airflow obstruction, such as bronchial asthma, is characterized by mucus hypersecretion, goblet cell hyperplasia(GCH), smooth muscle hypertrophy, and inflammatory cells infiltration. In fatal asthma patients, one distinct findings is mucus hypersecretion due to GCH. However, the mechanisms of GCH in these hypersecretory diseases remain still unknown. In this study, a rat model was rapidly induced with GCH by instillation of $TiO_2$, intratracheally. We intend to confirm GCH and association of concomitant inflammatory cells infiltration and to observe the effect of potent antiinflammatory agent, that is dexamethasone, on GCH with inflammatory cells. Methods : Twenty-one 8-weeks-old male Sprague-Dawley rats were divided into three groups. Endotoxinfree water was instilled intratracheally in group 1(control) ; $TiO_2$, was instilled in the group 2 ; and dexamethasone was injected intraperitoneally to group 3 before $TiO_2$ instillation. After 120 hours, all rats were sacrificed, and trachea, bronchi, and lungs were resected respectively. These tissues were made as paraffin blocks and stained as PAS for goblet cells and Luna stain for eosinophils. We calculated the ratio of goblet cell to respiratory epithelium and number of infiltrated eosinophils from each tissue. Results : (1) Fraction of goblet cells was significantly increased in group 2 than in group 1 in the trachea and in the main bronchus. (10.19$\pm$11.33% vs 4.09$\pm$8.28%, p<0.01 and 34.09$\pm$23.91% vs 3.61$\pm$4.84%, p<0.01, respectively). (2) Eosinophils were significantly increased in the airway of group 2 than that of group 1. (5.43$\pm$3.84% vs 0.17$\pm$0.47 in trachea and 47.71$\pm$16.91 vs 2.71$\pm$1.96 in main bronchi). (3) There was a positive correlation between goblet cells and eosinophils(r=0.719, p=0.001). (4) There was significant difference in the decrease of goblet cells after dexamethasone injection between group 2 and group 3 (p<0.01). Also, infiltration of eosinophils was suppressed by dexamethasone. Conclusion : We made an animal model of $TiO_2$-induced goblet cell hyperplasia. GCH was observed mainly in the main bronchi with concomitant eosinophilic infiltration. Both goblet cell hyperplasia and eosinophilic infiltration were suppressed by dexamethasone. This animal model may serve as a useful tool in understanding of the mechanism of GCH in chronic airway diseases.

• Proceedings of the KOR-BRONCHOESO Conference
• /
• 1981.05a
• /
• pp.7-8
• /
• 1981

Authors observed clinically 34 cases of the corrosive esophagitis caused by various corrosive agents at Kyung Hee University Hospital from Aug. 1978 to Dec. 1980. The results obtained were as follows; 1. Among the 34 patients, male was 19 (55.9%) and female 15(44.1%). Most frequently found age was 3rd decade. 2. 18 cases(52.9%) came to the hospital within 24 hours after ingestion of the agents, and 13 cases(38.2%) within 2 to 7 days. 3. Seasonal distribution showed most frequently in spring(35.3%). 4. The moment of the accident was suicidal attempt in 27 cases(79.4%) and misdrinking in 7 cases(20.6%). 5. Acetic acid was a most commonly used agent, showing 23 cases(67.6%), lye and insecticides were next in order. 6. Common chief complaints were swallowing difficulty and sore throat. 7. The average hospital days was 14.8 days. 8. Esophagogram was performed between 3 to 7 days after ingestion in 13 cases(38.2 %), findings were constrictions on the 1st narrowing portion in 4 cases(30.8%) and within normal limits in 3 cases(23.1%). 9. Esophagoscopy was performed in 31 cases(91.2%) between 2 to 7 days after ingestion, which revealed edema and coating on entrance of the esophagus in 9 cases (29.0 %). Diffuse edema on entire length of the esophagus and within normal limits were next in order. 10. Laboratory results were as follows: Anemia was in 1 cases(2.9%), leukocytosis. in 21 cases (61.8%), increase ESR in 9 cases (26.5%), markedly increased BUN and creatinine in 3 cases (8.8%), and hypokalemia in 1 cases(2.9%). Proteinuria in 10 cases(29.4%) hematuria in 4 cases(l1.8%), and coca cola urine in 3 cases (8.8%). 11. Associated diseases were 3 cases(8.8%) of cancer, 1 cases (2.9%) of diabetes mellitus, and 1 cases(2.9%) of manic depressive illness. 12. Various treatment was given: Esophageal and gastric washing in 23 cases(67.6%) for the emergent treatment, antibiotics in 32 cases(94.1%), steroids in 30 cases(88.2%), bougienation in 5 cases(14.7%), hemodialysis in 1 case(2.9%), and partial esophagectomy with gastrostomy and gastroileal anastomosis in 1 cases(2.9%). 13. Serious complications were observed in 9 cases (26.5%), consisted of 6 cases(17.6%) of esophageal stricture, 1 cases(2.9%), of aute renal failure, 1 cases (2.9%) of pneu momediastinum with pneumonia, and 1 cases (2.9%) of pneumonia.

• Tuberculosis and Respiratory Diseases
• /
• v.58 no.6
• /
• pp.562-569
• /
• 2005

Background : The severity scoring system is useful for predicting the outcome of critically ill patients. However, the system is quite complicated and cost-ineffective. Simple serologic markers have been proposed to predict the outcome, which include troponin-I, lactate and C-reactive protein(CRP). The aim of this study was to evaluate the prognostic values of troponin-I, lactate and CRP in critically ill non-cardiac patients. Methods : From September 2003 to June 2004, 139 patients(Age: $63.3{\pm}14.7$, M:F = 88:51), who were admitted to the MICU with non-cardiac critical illness at Gyeongsang National University Hospital, were enrolled in this study. This study evaluated the severity of the illness and the multi-organ failure score (Acute Physiologic and Chronic Health EvaluationII, Simplified Acute Physiologic ScoreII and Sequential Organ Failure Assessment) and measured the troponin-I, lactate and CRP within 24 hours after admission in the MICU. Each value in the survivors and non-survivors was compared at the 10th and 30th day after ICU admission. The mortality rate was compared at 10th and 30th day in normal and abnormal group. In addition, the correlations between each value and the severity score were assessed. Results : There were significantly higher troponin-I and CRP levels, not lactate, in the non-survivors than in the survivors at 10th day($1.018{\pm}2.58ng/ml$, $98.48{\pm}69.24mg/L$ vs. $4.208{\pm}10.23ng/ml$, $137.69{\pm}70.18mg/L$) (p<0.05). There were significantly higher troponin-I, lactate and CRP levels in the non-survivors than in the survivors on the 30th day ($0.99{\pm}2.66ng/ml$, $8.02{\pm}9.54ng/dl$, $96.87{\pm}68.83mg/L$ vs. $3.36{\pm}8.74ng/ml$, $15.42{\pm}20.57ng/dl$, $131.28{\pm}71.23mg/L$) (p<0.05). The mortality rate was significantly higher in the abnormal group of troponin-I, lactate and CRP than in the normal group of troponin-I, lactate and CRP at 10th day(28.1%, 31.6%, 18.9% vs. 11.0%, 15.8 %, 0%) and 30th day(38.6%, 47.4%, 25.8% vs. 15.9%, 21.7%, 14.3%) (p<0.05). Troponin-I and lactate were significantly correlated with the SAPS II score($r^2=0.254$, 0.365, p<0.05). Conclusion : Measuring the troponin-I, lactate and CRP levels upon admission may be useful for predicting the outcome of critically ill non-cardiac patients.

• Journal of Intelligence and Information Systems
• /
• v.26 no.2
• /
• pp.105-129
• /
• 2020

This study uses corporate data from 2012 to 2018 when K-IFRS was applied in earnest to predict default risks. The data used in the analysis totaled 10,545 rows, consisting of 160 columns including 38 in the statement of financial position, 26 in the statement of comprehensive income, 11 in the statement of cash flows, and 76 in the index of financial ratios. Unlike most previous prior studies used the default event as the basis for learning about default risk, this study calculated default risk using the market capitalization and stock price volatility of each company based on the Merton model. Through this, it was able to solve the problem of data imbalance due to the scarcity of default events, which had been pointed out as the limitation of the existing methodology, and the problem of reflecting the difference in default risk that exists within ordinary companies. Because learning was conducted only by using corporate information available to unlisted companies, default risks of unlisted companies without stock price information can be appropriately derived. Through this, it can provide stable default risk assessment services to unlisted companies that are difficult to determine proper default risk with traditional credit rating models such as small and medium-sized companies and startups. Although there has been an active study of predicting corporate default risks using machine learning recently, model bias issues exist because most studies are making predictions based on a single model. Stable and reliable valuation methodology is required for the calculation of default risk, given that the entity's default risk information is very widely utilized in the market and the sensitivity to the difference in default risk is high. Also, Strict standards are also required for methods of calculation. The credit rating method stipulated by the Financial Services Commission in the Financial Investment Regulations calls for the preparation of evaluation methods, including verification of the adequacy of evaluation methods, in consideration of past statistical data and experiences on credit ratings and changes in future market conditions. This study allowed the reduction of individual models' bias by utilizing stacking ensemble techniques that synthesize various machine learning models. This allows us to capture complex nonlinear relationships between default risk and various corporate information and maximize the advantages of machine learning-based default risk prediction models that take less time to calculate. To calculate forecasts by sub model to be used as input data for the Stacking Ensemble model, training data were divided into seven pieces, and sub-models were trained in a divided set to produce forecasts. To compare the predictive power of the Stacking Ensemble model, Random Forest, MLP, and CNN models were trained with full training data, then the predictive power of each model was verified on the test set. The analysis showed that the Stacking Ensemble model exceeded the predictive power of the Random Forest model, which had the best performance on a single model. Next, to check for statistically significant differences between the Stacking Ensemble model and the forecasts for each individual model, the Pair between the Stacking Ensemble model and each individual model was constructed. Because the results of the Shapiro-wilk normality test also showed that all Pair did not follow normality, Using the nonparametric method wilcoxon rank sum test, we checked whether the two model forecasts that make up the Pair showed statistically significant differences. The analysis showed that the forecasts of the Staging Ensemble model showed statistically significant differences from those of the MLP model and CNN model. In addition, this study can provide a methodology that allows existing credit rating agencies to apply machine learning-based bankruptcy risk prediction methodologies, given that traditional credit rating models can also be reflected as sub-models to calculate the final default probability. Also, the Stacking Ensemble techniques proposed in this study can help design to meet the requirements of the Financial Investment Business Regulations through the combination of various sub-models. We hope that this research will be used as a resource to increase practical use by overcoming and improving the limitations of existing machine learning-based models.

• The Korean Journal of Nuclear Medicine Technology
• /
• v.25 no.2
• /
• pp.35-40
• /
• 2021

Purpose Glomerular filtration rate(GFR) is an important indicator for the diagnosis, treatment, and follow-up of kidney disease and is also used by healthy individuals for drug use and evaluating kidney function in donors. The gold standard method of the GFR test is to measure by continuously injecting the inulin which is extrinsic marker, but it takes a long time and the test method is complicated. so, the method of measuring the serum concentration of creatinine is used. Estimated glomerular filtration rate (eGFR) is used instead. However, creatinine is known to be affected by age, gender, muscle mass, etc. eGFR formulas that are currently used include the Cockroft-Gault formula, the modification of diet in renal disease (MDRD) formula, and the chronic kidney disease epidemilogy collaboration (CKD-EPI) formula for adults. For children, the Schwartz formula is used. Measurement of GFR using ⁵¹Cr-EDTA (diethylenetriamine tetraacetic acid), ^99mTc-DTPA (diethylenetriamine pentaacetic acid) can replace inulin and is currently in use. Therefore, We compared the GFR measured using ^99mTc-DTPA with the eGFR using CKD-EPI formula. Materials and Methods For 200 kidney transplant donors who visited Asan medical center.(96 males, 104 females, 47.3 years ± 12.7 years old) GFR was measured using plasma(Two-plasma-sample-method, TPSM) obtained by intravenous administration of ^99mTc-DTPA(0.5mCi, 18.5 MBq). eGFR was derived using CKD-EPI formula based on serum creatinine concentration. Results GFR average measured using ^99mTc-DTPA for 200 kidney transplant donors is 97.27±19.46(ml/min/1.73m²), and the eGFR average value using the CKD-EPI formula is 96.84±17.74(ml/min/1.73m²), The concentration of serum creatinine is 0.84±0.39(mg/dL). Regression formula of ^99mTc-DTPA GFR for serum creatinine-based eGFR was Y = 0.5073X + 48.186, and the correlation coefficient was 0.698 (P<0.01). Difference (%) was 1.52±18.28. Conclusion The correlation coefficient between the ^99mTc-DTPA and the eGFR derived on serum creatinine concentration was confirmed to be moderate. This is estimated that eGFR is affected by external factors such as age, gender, and muscle mass and use of formulas made for kidney disease patients. By using ^99mTc-DTPA, we can provide reliable GFR results, which is used for diagnosis, treatment and observation of kidney disease, and kidney evaluation of kidney transplant patients.

• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.205-225
• /
• 2018

Convolutional Neural Network (ConvNet) is one class of the powerful Deep Neural Network that can analyze and learn hierarchies of visual features. Originally, first neural network (Neocognitron) was introduced in the 80s. At that time, the neural network was not broadly used in both industry and academic field by cause of large-scale dataset shortage and low computational power. However, after a few decades later in 2012, Krizhevsky made a breakthrough on ILSVRC-12 visual recognition competition using Convolutional Neural Network. That breakthrough revived people interest in the neural network. The success of Convolutional Neural Network is achieved with two main factors. First of them is the emergence of advanced hardware (GPUs) for sufficient parallel computation. Second is the availability of large-scale datasets such as ImageNet (ILSVRC) dataset for training. Unfortunately, many new domains are bottlenecked by these factors. For most domains, it is difficult and requires lots of effort to gather large-scale dataset to train a ConvNet. Moreover, even if we have a large-scale dataset, training ConvNet from scratch is required expensive resource and time-consuming. These two obstacles can be solved by using transfer learning. Transfer learning is a method for transferring the knowledge from a source domain to new domain. There are two major Transfer learning cases. First one is ConvNet as fixed feature extractor, and the second one is Fine-tune the ConvNet on a new dataset. In the first case, using pre-trained ConvNet (such as on ImageNet) to compute feed-forward activations of the image into the ConvNet and extract activation features from specific layers. In the second case, replacing and retraining the ConvNet classifier on the new dataset, then fine-tune the weights of the pre-trained network with the backpropagation. In this paper, we focus on using multiple ConvNet layers as a fixed feature extractor only. However, applying features with high dimensional complexity that is directly extracted from multiple ConvNet layers is still a challenging problem. We observe that features extracted from multiple ConvNet layers address the different characteristics of the image which means better representation could be obtained by finding the optimal combination of multiple ConvNet layers. Based on that observation, we propose to employ multiple ConvNet layer representations for transfer learning instead of a single ConvNet layer representation. Overall, our primary pipeline has three steps. Firstly, images from target task are given as input to ConvNet, then that image will be feed-forwarded into pre-trained AlexNet, and the activation features from three fully connected convolutional layers are extracted. Secondly, activation features of three ConvNet layers are concatenated to obtain multiple ConvNet layers representation because it will gain more information about an image. When three fully connected layer features concatenated, the occurring image representation would have 9192 (4096+4096+1000) dimension features. However, features extracted from multiple ConvNet layers are redundant and noisy since they are extracted from the same ConvNet. Thus, a third step, we will use Principal Component Analysis (PCA) to select salient features before the training phase. When salient features are obtained, the classifier can classify image more accurately, and the performance of transfer learning can be improved. To evaluate proposed method, experiments are conducted in three standard datasets (Caltech-256, VOC07, and SUN397) to compare multiple ConvNet layer representations against single ConvNet layer representation by using PCA for feature selection and dimension reduction. Our experiments demonstrated the importance of feature selection for multiple ConvNet layer representation. Moreover, our proposed approach achieved 75.6% accuracy compared to 73.9% accuracy achieved by FC7 layer on the Caltech-256 dataset, 73.1% accuracy compared to 69.2% accuracy achieved by FC8 layer on the VOC07 dataset, 52.2% accuracy compared to 48.7% accuracy achieved by FC7 layer on the SUN397 dataset. We also showed that our proposed approach achieved superior performance, 2.8%, 2.1% and 3.1% accuracy improvement on Caltech-256, VOC07, and SUN397 dataset respectively compare to existing work.