Extraction of Nucleus Region from Stained Cancer Image by Using Self-Growing Network

Yukiteru Zukeyama,Itaru Nagayama,Tomio Takara 대한전자공학회 1997 ITC-CSCC :International Technical Conference on Ci Vol.- No.-

Intelligent Traffic Prediction by Multi-sensor Fusion using Multi-threaded Machine Learning

Swe Sw Aung,Itaru Nagayama,Shiro Tamaki 대한전자공학회 2016 IEIE Transactions on Smart Processing & Computing Vol.5 No.6

Estimation and analysis of traffic jams plays a vital role in an intelligent transportation system and advances safety in the transportation system as well as mobility and optimization of environmental impact. For these reasons, many researchers currently mainly focus on the brilliant machine learning-based prediction approaches for traffic prediction systems. This paper primarily addresses the analysis and comparison of prediction accuracy between two machine learning algorithms: Naïve Bayes and K-Nearest Neighbor (K-NN). Based on the fact that optimized estimation accuracy of these methods mainly depends on a large amount of recounted data and that they require much time to compute the same function heuristically for each action, we propose an approach that applies multi-threading to these heuristic methods. It is obvious that the greater the amount of historical data, the more processing time is necessary. For a real-time system, operational response time is vital, and the proposed system also focuses on the time complexity cost as well as computational complexity. It is experimentally confirmed that K-NN does much better than Naïve Bayes, not only in prediction accuracy but also in processing time. Multithreading- based K-NN could compute four times faster than classical K-NN, whereas multithreading - based Naïve Bayes could process only twice as fast as classical Bayes.

Plurality Rule–based Density and Correlation Coefficient–based Clustering for K-NN

Swe Swe Aung,Itaru Nagayama,Shiro Tamaki 대한전자공학회 2017 IEIE Transactions on Smart Processing & Computing Vol.6 No.3

k-nearest neighbor (K-NN) is a well-known classification algorithm, being feature space–based on nearest-neighbor training examples in machine learning. However, K-NN, as we know, is a lazy learning method. Therefore, if a K-NN–based system very much depends on a huge amount of history data to achieve an accurate prediction result for a particular task, it gradually faces a processing-time performance-degradation problem. We have noticed that many researchers usually contemplate only classification accuracy. But estimation speed also plays an essential role in realtime prediction systems. To compensate for this weakness, this paper proposes correlation coefficient–based clustering (CCC) aimed at upgrading the performance of K-NN by leveraging processing-time speed and plurality rule–based density (PRD) to improve estimation accuracy. For experiments, we used real datasets (on breast cancer, breast tissue, heart, and the iris) from the University of California, Irvine (UCI) machine learning repository. Moreover, real traffic data collected from Ojana Junction, Route 58, Okinawa, Japan, was also utilized to lay bare the efficiency of this method. By using these datasets, we proved better processing-time performance with the new approach by comparing it with classical K-NN. Besides, via experiments on realworld datasets, we compared the prediction accuracy of our approach with density peaks clustering based on K-NN and principal component analysis (DPC-KNN-PCA).

A High-performance Classifier from K-dimensional Tree-based Dual-kNN

Swe Swe Aung,Itaru Nagayama,Shiro Tamaki 대한전자공학회 2018 IEIE Transactions on Smart Processing & Computing Vol.7 No.3

The k-nearest neighbors (kNN) method is highly effective in many application areas. Conceptually, its other good properties are simplicity and ease of understanding. However, according to measurements of the performance of algorithms based on three considerations (simplicity, processing time, and prediction power), the classic kNN algorithm lacks high-speed computation as well as maintenance of high accuracy for different values of k. The k-nearest neighbors algorithm is still influenced by varying k values and high variance in the training data set. Prediction accuracy diminishes when k approaches larger values. To overcome these issues, this paper introduces a k-dimensional (kd)-tree–based dual-kNN approach that concentrates on two properties to maintain classification accuracy at different k values and that also upgrades processing time performance. By conducting experiments on real data sets and comparing this algorithm with two other algorithms (dual-kNN and classic kNN), it was experimentally confirmed that the kdtree–based dual-kNN is a more effective and robust approach for classification than pure dual-kNN and classic kNN.

An Adaptive Morphological Operation for High-performance Weather Image Processing

Swe Swe Aung,Itaru Nagayama,Shiro Tamaki 대한전자공학회 2018 IEIE Transactions on Smart Processing & Computing Vol.7 No.6

Morphological operations have been an integral part of the enhancement of digital imaging programs, especially for filtering noise to improve the quality of images by utilizing the two most basic morphological operations: erosion and dilation. The main role of dilation is to fill the defined region in an image with pixels, whereas erosion removes pixels from the region. As we know, the methods of erosion followed by dilation, or dilation followed by erosion, are indeed attractive approaches amongst researchers who deal with filtering noise problems. However, these approaches need more computational time and have a high-percentage chance of losing essential pixel area. To cover these issues, this paper introduces a new approach called an adaptive morphological operation to boost the performance of image enhancement. Based on 2011, 2013, 2015, and 2016 weather image datasets collected from the WITH radar installed on the rooftop of the Information Engineering building, University of the Ryukyus, the experimental results confirm that the proposed approach is more efficient than the conventional approaches.

Dual-kNN for a Pattern Classification Approach

Swe Swe Aung,Itaru Nagayama,Shiro Tamaki 대한전자공학회 2017 IEIE Transactions on Smart Processing & Computing Vol.6 No.5

Classification is a process of discovering and categorizing objects from large data storage that have similar characteristics, properties, and patterns. One of the most widely used classification methods in machine learning is the k-nearest neighbors (k-NN) algorithm. The unique property of k-NN that appeals to researchers is its simplicity, so it can be applied successfully over a wide field. However, according to measurement of the performance of an algorithm based on three considerations (simplicity, processing time, and prediction power), the k-NN algorithm lacks highspeed computation and maintenance of high accuracy for different K values. In other words, k-NN is a heuristic classification approach. Besides, the prediction accuracy fades away whenever K approaches larger values. To overcome these issues, this paper presents a dual-kNN that concentrates on two properties to keep up the accuracy at different K values and upgrade processing time performance. By conducting experiments on real datasets and comparing this algorithm with k-NN, it was also confirmed that the new dual-kNN is an effective and robust approach to classification.

Traffic Flow Estimation System using a Hybrid Approach

Swe Sw Aung,Itaru Nagayama,Shiro Tamaki 대한전자공학회 2017 IEIE Transactions on Smart Processing & Computing Vol.6 No.4

Nowadays, as traffic jams are a daily elementary problem in both developed and developing countries, systems to monitor, predict, and detect traffic conditions are playing an important role in research fields. Comparing them, researchers have been trying to solve problems by applying many kinds of technologies, especially roadside sensors, which still have some issues, and for that reason, any one particular method by itself could not generate sufficient traffic prediction results. However, these sensors have some issues that are not useful for research. Therefore, it may not be best to use them as stand-alone methods for a traffic prediction system. On that note, this paper mainly focuses on predicting traffic conditions based on a hybrid prediction approach, which stands on accuracy comparison of three prediction models: multinomial logistic regression, decision trees, and support vector machine (SVM) classifiers. This is aimed at selecting the most suitable approach by means of integrating proficiencies from these approaches. It was also experimentally confirmed, with test cases and simulations that showed the performance of this hybrid method is more effective than individual methods.

Investigation into Tolerance of Mislabeling when Classifying Patterns with Dual-kNN

Swe Swe Aung,Itaru Nagayama,Shiro Tamaki 대한전자공학회 2018 IEIE Transactions on Smart Processing & Computing Vol.7 No.6

As we know, machine learning algorithms are powerful tools for a variety of application domains, giving widely divergent dimensions, such as reliability, precision, robustness, high-speed solutions, etc. Likewise, the other critical dimension that a well-designed learning algorithm should occupy is strength against unpredictable and phenomenal noise. For this critical dimension, we introduce a new approach, dual k-nearest neighbors (dual-kNN), to investigate the tolerance level for mislabeling based on different injected-noise levels. Literally, dual-kNN is a reborn algorithm of k-nearest neighbors (k-NN) aiming to reduce the influence of a steady decrease in prediction accuracy over increasing k values. What is more, dual-kNN is proven to have higher classification accuracy in many application domains. For the primary goal of this paper, we mainly emphasize investigating dual-kNN’s resistance level to mislabeled classes. Provably, our empirical experimentations describe how dual-kNN has a higher resistance level to mislabeling than normal k-NN, density-based kNN, and logistic regression, for noise levels of up to 50%. The practical datasets applied within this paper are medical data files from the University of California, Irvine (UCI) Machine Learning Repository.

Noise-tolerance Investigation into Dual-kNN Pattern Classification

Swe Swe Aung,Itaru Nagayama,Shiro Tamaki 대한전자공학회 2018 IEIE Transactions on Smart Processing & Computing Vol.7 No.3

The performance of an algorithm is usually measured in three dimensions (simplicity, processing time, and prediction power). In addition, we should take into account the noise resistance level in those measures. For this reason, this paper focuses on investigating the noisetolerance level of dual k-nearest neighbors (dual-kNN) primarily based on five noisy medical diagnosis problems. Literally, dual-kNN is a reborn version of the k-nearest neighbors (k-NN) algorithm with a new observation idea in the classification process with a collaborative effort between the first and second nearest neighbors of an observed instance. It was recently proven that dual-kNN has high prediction accuracy for a variety of real-world data sets, especially so in unbiased data sets. Thus, in this report, not only the prediction accuracy of dual-kNN is compared with normal k-NN, logistic regression, and the neural network, but we additionally investigate the noise tolerance in the aforementioned approaches. The practical data sets applied in this paper are medical data files from the University of California, Irvine, Machine Learning Repository. In this report, the new approach to dual-kNN commences with better prediction accuracy, and higher noise resistance is presented, in comparison with normal k-NN, logistic regression, and neural networks.

Short-term Prediction of Localized Heavy Rain from Radar Imaging and Machine Learning

Swe Swe Aung,Yu Senaha,Shin Ohsawa,Itaru Nagayama,Shiro Tamaki 대한전자공학회 2018 IEIE Transactions on Smart Processing & Computing Vol.7 No.2

Heavy rainfall has frequently caused serious flooding and landslides, increasing traffic delays in most parts of the world. Consequently, the people in areas battered by heavy rainfall face many hardships. Thus, the negative effects of torrential rainfall always remind researchers to keep seeking the ways to prevent such damage. Therefore, we designed a system for short-term prediction of localized heavy downpours by using radar images coupled with a machine learning method. Here, we introduce a new approach, named dual k-nearest neighbor (dual-kNN), for shortterm rainfall prediction by upgrading the ordinary classification routines of classical k-nearest neighbors (k-NN). dual-kNN is able to maintain highly robust classification of various K values with an advanced simple dual consideration, where observation of a targeted object can be found not only in the specified region but also in other related regions. We conducted experimentations using 2011, 2013, and 2014 data sets collected from the WITH small-dish aviation radar installed on the rooftop of Information Engineering, University of the Ryukyus. Then, we compared the prediction accuracy of our new approach with classical k-NN. It was experimentally confirmed with test cases and simulations that the performance of dual-kNN is more effective than classical k-NN.