• Journal of Intelligence and Information Systems
• /
• v.17 no.4
• /
• pp.109-130
• /
• 2011

One of the major problems in the area of data mining is the size of the data, as most data set has huge volume these days. Streams of data are normally accumulated into data storages or databases. Transactions in internet, mobile devices and ubiquitous environment produce streams of data continuously. Some data set are just buried un-used inside huge data storage due to its huge size. Some data set is quickly lost as soon as it is created as it is not saved due to many reasons. How to use this large size data and to use data on stream efficiently are challenging questions in the study of data mining. Stream data is a data set that is accumulated to the data storage from a data source continuously. The size of this data set, in many cases, becomes increasingly large over time. To mine information from this massive data, it takes too many resources such as storage, money and time. These unique characteristics of the stream data make it difficult and expensive to store all the stream data sets accumulated over time. Otherwise, if one uses only recent or partial of data to mine information or pattern, there can be losses of valuable information, which can be useful. To avoid these problems, this study suggests a method efficiently accumulates information or patterns in the form of rule set over time. A rule set is mined from a data set in stream and this rule set is accumulated into a master rule set storage, which is also a model for real-time decision making. One of the main advantages of this method is that it takes much smaller storage space compared to the traditional method, which saves the whole data set. Another advantage of using this method is that the accumulated rule set is used as a prediction model. Prompt response to the request from users is possible anytime as the rule set is ready anytime to be used to make decisions. This makes real-time decision making possible, which is the greatest advantage of this method. Based on theories of ensemble approaches, combination of many different models can produce better prediction model in performance. The consolidated rule set actually covers all the data set while the traditional sampling approach only covers part of the whole data set. This study uses a stock market data that has a heterogeneous data set as the characteristic of data varies over time. The indexes in stock market data can fluctuate in different situations whenever there is an event influencing the stock market index. Therefore the variance of the values in each variable is large compared to that of the homogeneous data set. Prediction with heterogeneous data set is naturally much more difficult, compared to that of homogeneous data set as it is more difficult to predict in unpredictable situation. This study tests two general mining approaches and compare prediction performances of these two suggested methods with the method we suggest in this study. The first approach is inducing a rule set from the recent data set to predict new data set. The seocnd one is inducing a rule set from all the data which have been accumulated from the beginning every time one has to predict new data set. We found neither of these two is as good as the method of accumulated rule set in its performance. Furthermore, the study shows experiments with different prediction models. The first approach is building a prediction model only with more important rule sets and the second approach is the method using all the rule sets by assigning weights on the rules based on their performance. The second approach shows better performance compared to the first one. The experiments also show that the suggested method in this study can be an efficient approach for mining information and pattern with stream data. This method has a limitation of bounding its application to stock market data. More dynamic real-time steam data set is desirable for the application of this method. There is also another problem in this study. When the number of rules is increasing over time, it has to manage special rules such as redundant rules or conflicting rules efficiently.

• Journal of Digital Convergence
• /
• v.18 no.11
• /
• pp.259-266
• /
• 2020

Many developments have been steadily carried out by researchers with applying knowledge-based expert system or machine learning algorithms to the financial field. In particular, it is now common to perform knowledge based system trading in using stock prices. Recently, deep learning technologies have been applied to real fields of stock trading marketplace as GPU performance and large scaled data have been supported enough. Especially, LSTM has been tried to apply to stock price prediction because of its compatibility for time series data. In this paper, we implement stock price prediction using LSTM. In modeling of LSTM, we propose a fitness combination of model parameters and activation functions for best performance. Specifically, we propose suitable selection methods of initializers of weights and bias, regularizers to avoid over-fitting, activation functions and optimization methods. We also compare model performances according to the different selections of the above important modeling considering factors on the real-world stock price data of global major companies. Finally, our experimental work brings a fitness method of applying LSTM model to stock price prediction.

• Journal of Adhesion and Interface
• /
• v.17 no.2
• /
• pp.49-55
• /
• 2016

In this study, we designed an environment friendly, water-based adhesive using the acrylic emulsion method as a replacement for solvent-based adhesives, which are most commonly used in layered laminates for flexible food packaging. We designed adhesives with different combinations of anionic, non-ionic, and phosphoric ester surfactants, and with different concentrations of chain transfer agent (CTA). We also examined the effect of the degree of cross-linking by synthesizing and comparing 8 test group adhesives with different types of functional monomers. Additionally, we synthesized 2 other test group pressure-sensitive adhesives (PSA) using styrene/alpha-methyl styrene/acrylic acid (SAA) semipolymer dispersing agents (with molecular weights of 13,000 g/mol and 8,600 g/mol, respectively) to replace the conventional surfactants. We evaluated whether the 10 test group pressure-sensitive adhesives met the basic physical property criteria required for flexible food packaging by carrying out a physical analysis of their glass transition temperature (Tg), particle size, adhesion, and molecular weight. In our test, 2 test group adhesives manufactured with the combination of anionic and non-ionic surfactants, CTA concentration of 0.2%, and functional monomers of hydroxyethyl acrylate (HEA) and glycidyl methacrylate (GMA) demonstrated molecular weight and flexibility suitable for flexible packaging, with low adhesiveness and small particle size.

• Magazine of the Korean Society of Agricultural Engineers
• /
• v.16 no.2
• /
• pp.3361-3394
• /
• 1974

The purpose of this thesis is to disclose some characteristics of water consumption in relation to the quantities of dry matters through the growing period for two statured varieties of paddy rice which are a tall statured variety and a short one, including the water consumption during seedling period, and to find out the various coefficients of evapotranspiration that are applicable for the water use of an expected yield of the two varieties. PAL-TAL, a tall statured variety, and TONG-lL, a short statured variety were chosen for this investigation. Experiments were performed in two consecutive periods, a seedling period and a paddy field period, In the investigation of seedling period, rectangular galvanized iron evapotranspirometers (91cm${\times}$85cm${\times}$65cm) were set up in a way of two levels (PAL-TAL and TONG-lL varieties) with two replications. A standard fertilization method was applied to all plots. In the experiment of paddy field period, evapotanspiration and evaporation were measured separately. For PAL-TAL variety, the evapotranspiration measurements of 43 plots of rectangular galvanized iron evapotranspirometer (91cm${\times}$85cm${\times}$65cm) and the evaporation measurements of 25 plots of rectangular galvanized iron evaporimeter (91cm${\times}$85cm${\times}$15cm) have been taken for seven years (1966 through 1972), and for TONG-IL variety, the evapotranspiration measurements of 19 plots and the evaporation measurements of 12 plots have been collected for two years (1971 through 1972) with five different fertilization levels. The results obtained from this investigation are summarized as follows: 1. Seedling period 1) The pan evaporation and evapotranspiration during seedling period were proved to have a highly significant correlation to solar radiation, sun shine hours and relative humidity. But they had no significant correlation to average temperature, wind velocity and atmospheric pressure, and were appeared to be negatively correlative to average temperature and wind velocity, and positively correlative to the atmospheric pressure, in a certain period. There was the highest significant correlation between the evapotranspiration and the pan evaporation, beyond all other meteorological factors considered. 2) The evapotranpiration and its coefficient for PAL-TAL variety were 194.5mm and 0.94∼1.21(1.05 in average) respectively, while those for TONG-lL variety were 182.8mm and 0.90∼1.10(0.99 in average) respectively. This indicates that the evapotranspiration for TONG-IL variety was 6.2% less than that for PAL-TAL variety during a seedling period. 3) The evapotranspiration ratio (the ratio of the evapotranspiration to the weight of dry matters) during the seedling period was 599 in average for PAL-TAL variety and 643 for TONG-IL variety. Therefore the ratio for TONG-IL was larger by 44 than that for PAL-TAL variety. 4) The K-values of Blaney and Criddle formula for PAL-TAL variety were 0.78∼1.06 (0.92 in average) and for TONG-lL variety 0.75∼0.97 (0.86 in average). 5) The evapotranspiration coefficient and the K-value of B1aney and Criddle formular for both PAL-TAL and TONG-lL varieties showed a tendency to be increasing, but the evapotranspiration ratio decreasing, with the increase in the weight of dry matters. 2. Paddy field period 1) Correlation between the pan evaporation and the meteorological factors and that between the evapotranspiration and the meteorological factors during paddy field period were almost same as that in case of the seedling period (Ref. to table IV-4 and table IV-5). 2) The plant height, in the same level of the weight of dry matters, for PAL-TAL variety was much larger than that for TONG-IL variety, and also the number of tillers per hill for PAL-TAL variety showed a trend to be larger than that for TONG-IL variety from about 40 days after transplanting. 3) Although there was a tendency that peak of leaf-area-index for TONG-IL variety was a little retarded than that for PAL-TAL variety, it appeared about 60∼80 days after transplanting. The peaks of the evapotranspiration coefficient and the weight of dry matters at each growth stage were overlapped at about the same time and especially in the later stage of growth, the leaf-area-index, the evapotranspiration coefficient and the weight of dry matters for TONG-IL variety showed a tendency to be larger then those for PAL-TAL variety. 4) The evaporation coefficient at each growth stage for TONG-IL and PAL-TALvarieties was decreased and increased with the increase and decrease in the leaf-area-index, and the evaporation coefficient of TONG-IL variety had a little larger value than that of PAL-TAL variety. 5) Meteorological factors (especially pan evaporation) had a considerable influence to the evapotranspiration, the evaporation and the transpiration. Under the same meteorological conditions, the evapotranspiration (ET) showed a increasing logarithmic function of the weight of dry matters (x), while the evaporation (EV) a decreasing logarithmic function of the weight of dry matters; 800kg/10a x 2000kg/10a, ET=al+bl logl0x (bl>0) EV=a2+b2 log10x (a2>0 b2<0) At the base of the weight of total dry matters, the evapotranspiration and the evaporation for TONG-IL variety were larger as much as 0.3∼2.5% and 7.5∼8.3% respectively than those of PAL-TAL variety, while the transpiration for PAL-TAL variety was larger as much as 1.9∼2.4% than that for TONG-IL variety on the contrary. At the base of the weight of rough rices the evapotranspiration and the transpiration for TONG-IL variety were less as much as 3.5% and 8.l∼16.9% respectively than those for PAL-TAL variety and the evaporation for TONG-IL was much larger by 11.6∼14.8% than that for PAL-TAL variety. 6) The evapotranspiration coefficient, the evaporation coefficient and the transpiration coefficient and the transpiration coefficient were affected by the weight of dry matters much more than by the meteorological conditions. The evapotranspiratioa coefficient (ETC) and the evaporation coefficient (EVC) can be related to the weight of dry matters (x) by the following equations: 800kg/10a x 2000kg/10a, ETC=a3+b3 logl0x (b3>0) EVC=a4+b4 log10x (a4>0, b4>0) At the base of the weights of dry matters, 800kg/10a∼2000kg/10a, the evapotranspiration coefficients for TONG-IL variety were 0.968∼1.474 and those for PAL-TAL variety, 0.939∼1.470, the evaporation coefficients for TONG-IL variety were 0.504∼0.331 and those for PAL-TAL variety, 0.469∼0.308, and the transpiration coefficients for TONG-IL variety were 0.464∼1.143 and those for PAL-TAL variety, 0.470∼1.162. 7) The evapotranspiration ratio, the evaporation ratio (the ratio of the evaporation to the weight of dry matters) and the transpiration ratio were highly affected by the meteorological conditions. And under the same meteorological condition, both the evapotranspiration ratio (ETR) and the evaporation ratio (EVR) showed to be a decreasing logarithmic function of the weight of dry matters (x) as follows: 800kg/10a x 2000kg/10a, ETR=a5+b5 logl0x (a5>0, b5<0) EVR=a6+b6 log10x (a6>0 b6<0) In comparison between TONG-IL and PAL-TAL varieties, at the base of the pan evaporation of 343mm and the weight of dry matters of 800∼2000kg/10a, the evapotranspiration ratios for TONG-IL variety were 413∼247, while those for PAL-TAL variety, 404∼250, the evaporation ratios for TONG-IL variety were 197∼38 while those for PAL-TAL variety, 182∼34, and the transpiration ratios for TONG-IL variety were 216∼209 while those for PAL-TAL variety, 222∼216 (Ref. to table IV-23, table IV-25 and table IV-26) 8) The accumulative values of evapotranspiration intensity and transpiration intensity for both PAL-TAL and TONG-IL varieties were almost constant in every climatic year without the affection of the weight of dry matters. Furthermore the evapotranspiration intensity appeared to have more stable at each growth stage. The peaks of the evapotranspiration intensity and transpiration intensity, for both TONG-IL and PAL-TAL varieties, appeared about 60∼70 days after transplanting, and the peak value of the former was 128.8${\pm}$0.7, for TONG-IL variety while that for PAL-TAL variety, 122.8${\pm}$0.3, and the peak value of the latter was 152.2${\pm}$1.0 for TONG-IL variety while that for PAL-TAL variety, 152.7${\pm}$1.9 (Ref.to table IV-27 and table IV-28) 9) The K-value in Blaney & Criddle formula was changed considerably by the meteorological condition (pan evaporation) and related to be a increasing logarithmic function of the weight of dry matters (x) for both PAL-TAL and TONG-L varieties as follows; 800kg/10a x 2000kg/10a, K=a7+b7 logl0x (b7>0) The K-value for TONG-IL variety was a little larger than that for PAL-TAL variety. 10) The peak values of the evapotranspiration coefficient and k-value at each growth stage for both TONG-IL and PAL-TAL varieties showed up about 60∼70 days after transplanting. The peak values of the former at the base of the weights of total dry matters, 800∼2000kg/10a, were 1.14∼1.82 for TONG-IL variety and 1.12∼1.80, for PAL-TAL variety, and at the base of the weights of rough rices, 400∼1000 kg/10a, were 1.11∼1.79 for TONG-IL variety and 1.17∼1.85 for PAL-TAL variety. The peak values of the latter, at the base of the weights of total dry matters, 800∼2000kg/10a, were 0.83∼1.39 for TONG-IL variety and 0.86∼1.36 for PAL-TAL variety and at the base of the weights of rough rices, 400∼1000kg/10a, 0.85∼1.38 for TONG-IL variety and 0.87∼1.40 for PAL-TAL variety (Ref. to table IV-18 and table IV-32) 11) The reasonable and practicable methods that are applicable for calculating the evapotranspiration of paddy rice in our country are to be followed the following priority a) Using the evapotranspiration coefficients based on an expected yield (Ref. to table IV-13 and table IV-18 or Fig. IV-13). b) Making use of the combination method of seasonal evapotranspiration coefficient and evapotranspiration intensity (Ref. to table IV-13 and table IV-27) c) Adopting the combination method of evapotranspiration ratio and evapotranspiration intensity, under the conditions of paddy field having a higher level of expected yield (Ref. to table IV-23 and table IV-27). d) Applying the k-values calculated by Blaney-Criddle formula. only within the limits of the drought year having the pan evaporation of about 450mm during paddy field period as the design year (Ref. to table IV-32 or Fig. IV-22).

• Korean Journal of Soil Science and Fertilizer
• /
• v.33 no.4
• /
• pp.221-233
• /
• 2000

Evaluation method of soil fertility by application of geographic information system (GIS) which includes landscape characteristics and soil map data was investigated from productivities of red pepper and tobacco grown on the fields with no fertilization. Total 131 fields experiments, 64 fields of red pepper and 67 fields of tobacco were conducted from 22 and 23 fields for red pepper and tobacco, respectively, located at Cheangweon and Eumseong counties in 1996, from 20 and 25 fields at Boeun and Goesan counties in 1997, and 22 and 19 fields at Jincheon and Chungju counties in 1998. All the experimental sites were selected on the basis of wide range of distribution in landscape and soil attributes. Dry weights and nutrients (N, P and K) uptakes by red pepper plant and tobacco leaves were considered as basic fertility of the soil (BFS). The BFS was estimated by twenty-five independent variables including 13 chemical properties and 12 GIS data. Twenty-five independent variables were classified by two groups, 15 quantitative variables and 10 qualitative variables, and were analyzed by multiple linear regression (MLR) of REG and GLM models of SAS. Dry weight of red pepper (DWRP) and dry weight of tobacco leaves (DWTL) every year showed high variations by five times in difference plots with minimum yield and maximum yield indicating the diverse soil fertility among the experimental fields. Evaluation for the BFS by the MLR including independent variables was better than that by simple regression showing gradual improvement by adding chemical properties, quantitative variables, and qualitative variables of the GIS. However the evaluation for the BFS by the MLR showed the better result for tobacco than red pepper. For example the variability in the DWTL by MLR was explained 34.2% by only chemical properties, 35.0% by adding quantitative variables, and 72.5% by adding both the quantitative and qualitative variables of the GIS compared with 21.7% by simple regression with $NO_3-N$ content in soil. Consequently, it is assumed that this approach by the MLR including both the quantitative and qualitative variables was available as an evaluation model of soil fertility for upland field.

• Journal of Intelligence and Information Systems
• /
• v.27 no.3
• /
• pp.139-156
• /
• 2021

The biggest reason for using a deep learning model in image classification is that it is possible to consider the relationship between each region by extracting each region's features from the overall information of the image. However, the CNN model may not be suitable for emotional image data without the image's regional features. To solve the difficulty of classifying emotion images, many researchers each year propose a CNN-based architecture suitable for emotion images. Studies on the relationship between color and human emotion were also conducted, and results were derived that different emotions are induced according to color. In studies using deep learning, there have been studies that apply color information to image subtraction classification. The case where the image's color information is additionally used than the case where the classification model is trained with only the image improves the accuracy of classifying image emotions. This study proposes two ways to increase the accuracy by incorporating the result value after the model classifies an image's emotion. Both methods improve accuracy by modifying the result value based on statistics using the color of the picture. When performing the test by finding the two-color combinations most distributed for all training data, the two-color combinations most distributed for each test data image were found. The result values were corrected according to the color combination distribution. This method weights the result value obtained after the model classifies an image's emotion by creating an expression based on the log function and the exponential function. Emotion6, classified into six emotions, and Artphoto classified into eight categories were used for the image data. Densenet169, Mnasnet, Resnet101, Resnet152, and Vgg19 architectures were used for the CNN model, and the performance evaluation was compared before and after applying the two-stage learning to the CNN model. Inspired by color psychology, which deals with the relationship between colors and emotions, when creating a model that classifies an image's sentiment, we studied how to improve accuracy by modifying the result values based on color. Sixteen colors were used: red, orange, yellow, green, blue, indigo, purple, turquoise, pink, magenta, brown, gray, silver, gold, white, and black. It has meaning. Using Scikit-learn's Clustering, the seven colors that are primarily distributed in the image are checked. Then, the RGB coordinate values of the colors from the image are compared with the RGB coordinate values of the 16 colors presented in the above data. That is, it was converted to the closest color. Suppose three or more color combinations are selected. In that case, too many color combinations occur, resulting in a problem in which the distribution is scattered, so a situation fewer influences the result value. Therefore, to solve this problem, two-color combinations were found and weighted to the model. Before training, the most distributed color combinations were found for all training data images. The distribution of color combinations for each class was stored in a Python dictionary format to be used during testing. During the test, the two-color combinations that are most distributed for each test data image are found. After that, we checked how the color combinations were distributed in the training data and corrected the result. We devised several equations to weight the result value from the model based on the extracted color as described above. The data set was randomly divided by 80:20, and the model was verified using 20% of the data as a test set. After splitting the remaining 80% of the data into five divisions to perform 5-fold cross-validation, the model was trained five times using different verification datasets. Finally, the performance was checked using the test dataset that was previously separated. Adam was used as the activation function, and the learning rate was set to 0.01. The training was performed as much as 20 epochs, and if the validation loss value did not decrease during five epochs of learning, the experiment was stopped. Early tapping was set to load the model with the best validation loss value. The classification accuracy was better when the extracted information using color properties was used together than the case using only the CNN architecture.

• 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.