• Proceedings of the Korean Society of Machine Tool Engineers Conference
• /
• 2004.10a
• /
• pp.280-285
• /
• 2004

In this study, we made a comparative study of backpropagation neural network and probabilistic neural network and bayesian classifier and perceptron as shape recognition algorithm of welding flaws. For this purpose, variables are applied the same to four algorithms. Here, feature variable is composed of time domain signal itself and frequency domain signal itself. Through this process, we comfirmed advantages/disadvantages of four algorithms and identified application methods of four algorithms.

• Korean Journal of Remote Sensing
• /
• v.35 no.6_1
• /
• pp.999-1009
• /
• 2019

The importance of high-resolution sea ice maps of the Arctic Ocean is increasing due to the possibility of pioneering North Pole Routes and the necessity of precise climate prediction models. In this study,sea ice classification algorithms for two deep learning models were examined using Sentinel-1 A/B SAR data to generate high-resolution sea ice classification maps. Based on current ice charts, three classes (Open Water, First Year Ice, Multi Year Ice) of training data sets were generated by Arctic sea ice and remote sensing experts. Ten sea ice classification algorithms were generated by combing two deep learning models (i.e. Simple CNN and Resnet50) and five cases of input bands including incident angles and thermal noise corrected HV bands. For the ten algorithms, analyses were performed by comparing classification results with ground truth points. A confusion matrix and Cohen's kappa coefficient were produced for the case that showed best result. Furthermore, the classification result with the Maximum Likelihood Classifier that has been traditionally employed to classify sea ice. In conclusion, the Convolutional Neural Network case, which has two convolution layers and two max pooling layers, with HV and incident angle input bands shows classification accuracy of 96.66%, and Cohen's kappa coefficient of 0.9499. All deep learning cases shows better classification accuracy than the classification result of the Maximum Likelihood Classifier.

• IE interfaces
• /
• v.12 no.4
• /
• pp.551-556
• /
• 1999

In this paper, we compare the performance of three data mining classification algorithms(neural network, decision tree, logistic regression) in consideration of various characteristics of categorical input and output data. $2^{4-1}$. 3 fractional factorial design is used to simulate the comparison situation where factors used are (1) the categorical ratio of input variables, (2) the complexity of functional relationship between the output and input variables, (3) the size of randomness in the relationship, (4) the categorical ratio of an output variable, and (5) the classification algorithm. Experimental study results indicate the following: decision tree performs better than the others when the relationship between output and input variables is simple while logistic regression is better when the other way is around; and neural network appears a better choice than the others when the randomness in the relationship is relatively large. We also use Taguchi design to improve the practicality of our study results by letting the relationship between the output and input variables as a noise factor. As a result, the classification accuracy of neural network and decision tree turns out to be higher than that of logistic regression, when the categorical proportion of the output variable is even.

• Journal of Korean Institute of Industrial Engineers
• /
• v.27 no.1
• /
• pp.47-53
• /
• 2001

In this paper, we compare the classification performances of both ensemble and clustering algorithms (Data Bagging, Variable Selection Bagging, Parameter Combining, Clustering) to logistic regression in consideration of various characteristics of input data. Four factors used to simulate the logistic model are (1) correlation among input variables (2) variance of observation (3) training data size and (4) input-output function. In view of the unknown relationship between input and output function, we use a Taguchi design to improve the practicality of our study results by letting it as a noise factor. Experimental study results indicate the following: When the level of the variance is medium, Bagging & Parameter Combining performs worse than Logistic Regression, Variable Selection Bagging and Clustering. However, classification performances of Logistic Regression, Variable Selection Bagging, Bagging and Clustering are not significantly different when the variance of input data is either small or large. When there is strong correlation in input variables, Variable Selection Bagging outperforms both Logistic Regression and Parameter combining. In general, Parameter Combining algorithm appears to be the worst at our disappointment.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.6
• /
• pp.1541-1549
• /
• 1997

Many classification algorithms require that training examples contain only discrete values. In order to use these algorithms when some attributes have continuous numeric values, the numeric attributes must be converted into discrete ones. This paper describes a new way of discretizing numeric values using information theory. Our method is context-sensitive in the sense that it takes into account the value of the target attribute. The amount of information each interval gives to the target attribute is measured using Hellinger divergence, and the interval boundaries are decided so that each interval contains as equal amount of information as possible. In order to compare our discretization method with some current discretization methods, several popular classification data sets are selected for experiment. We use back propagation algorithm and ID3 as classification tools to compare the accuracy of our discretization method with that of other methods.

• Journal of Sensor Science and Technology
• /
• v.17 no.6
• /
• pp.406-413
• /
• 2008

We have been implemented simple statistical pattern recognition methods for harmful gases classification using gas sensors array fabricated by MEMS (Micro Electro Mechanical System) technology. The performance of pattern recognition method as a gas classifier is highly dependent on the choice of pre-processing techniques for sensor and sensors array signals and optimal classification algorithms among the various classification techniques. We carried out pre-processing for each sensor's signal as well as sensors array signals to extract features for each gas. We adapted simple statistical pattern recognition algorithms, which were PCA (Principal Component Analysis) for visualization of patterns clustering and MLR (Multi-Linear Regression) for real-time system implementation, to classify harmful gases. Experimental results of adapted pattern recognition methods with pre-processing techniques have been shown good clustering performance and expected easy implementation for real-time sensing system.

• Journal of Korean Institute of Industrial Engineers
• /
• v.38 no.4
• /
• pp.276-284
• /
• 2012

Classification algorithms generally assume that the data is complete. However, missing values are common in real data sets due to various reasons. In this paper, we propose to use locally linear reconstruction (LLR) for missing value imputation to improve the classification performance when missing values exist. We first investigate how much missing values degenerate the classification performance with regard to various missing ratios. Then, we compare the proposed missing value imputation (LLR) with three well-known single imputation methods over three different classifiers using eight data sets. The experimental results showed that (1) any imputation methods, although some of them are very simple, helped to improve the classification accuracy; (2) among the imputation methods, the proposed LLR imputation was the most effective over all missing ratios, and (3) when the missing ratio is relatively high, LLR was outstanding and its classification accuracy was as high as the classification accuracy derived from the compete data set.

• Journal of Computing Science and Engineering
• /
• v.5 no.2
• /
• pp.150-160
• /
• 2011

Automatic text classification has a long history and many studies have been conducted in this field. In particular, many machine learning algorithms and information retrieval techniques have been applied to text classification tasks. Even though much technical progress has been made in text classification, there is still room for improvement in text classification. In this paper, we will discuss remaining issues in improving text classification. In this paper, three improvement issues are presented including automatic training data generation, noisy data treatment and term weighting and indexing, and four actual studies and their empirical results for those issues are introduced. First, the semi-supervised learning technique is applied to text classification to efficiently create training data. For effective noisy data treatment, a noisy data reduction method and a robust text classifier from noisy data are developed as a solution. Finally, the term weighting and indexing technique is revised by reflecting the importance of sentences into term weight calculation using summarization techniques.

• Journal of Information Processing Systems
• /
• v.3 no.2
• /
• pp.73-81
• /
• 2007

In this paper, we introduce a domain-independent classification framework based on both k-nearest neighbor and Naive Bayesian classification algorithms. The architecture of our system is simple and modularized in that each sub-module of the system could be changed or improved efficiently. Moreover, it provides various feature selection mechanisms to be applied to optimize the general-purpose classifiers for a specific domain. As for the enhanced classification performance, our system provides conditional probability boosting (CPB) mechanism which could be used in various domains. In the mood classification domain, our optimized framework using the CPB algorithm showed 1% of improvement in precision and 2% in recall compared with the baseline.

• The Journal of the Acoustical Society of Korea
• /
• v.16 no.1E
• /
• pp.3-8
• /
• 1997

Three new classification methods for multi temporal images are proposed. They are named as a likelihood addition method, a likelihood majority method and a Dempster-Shafer's rule method. Basic strategies using these methods are to calculate likelihoods for each temporal data and to combine obtained likelihoods for final classification. These three methods use different combining algorithms. From classification experiments, following results were obtained. The method based on Dempster-Shafer's rule of combination showed about 12% improvement of classification accuracies compared to a conventional method. This method needed about 16% more processing times than that of a conventional method. The other two proposed method showed 1% to 5% increase of classification accuracies. However processing times of these two proposed method showed 1% to 5% increase of classification accuracies. However processing times of these two methods are almost the same with that of a conventional method. Among the newly proposed three methods, the Dempster-Shafer's rule method showed the highest classification accuracies with more processing time than those of other methods.