• Information Systems Review
• /
• v.24 no.1
• /
• pp.1-19
• /
• 2022

Social media-based communication has become crucial part of our personal and official lives. Therefore, it is no surprise that social media sentiment analysis has emerged an important way of detecting potential customers' sentiment trends for all kinds of companies. However, social media sentiment analysis suffers from huge number of sentiment features obtained in the process of conducting the sentiment analysis. In this sense, this study proposes a novel method by using Bayesian Network. In this model MBFS (Markov Blanket-based Feature Selection) is used to reduce the number of sentiment features. To show the validity of our proposed model, we utilized online review data from Yelp, a famous social media about restaurant, bars, beauty salons evaluation and recommendation. We used a number of benchmarking feature selection methods like correlation-based feature selection, information gain, and gain ratio. A number of machine learning classifiers were also used for our validation tasks, like TAN, NBN, Sons & Spouses BN (Bayesian Network), Augmented Markov Blanket. Furthermore, we conducted Bayesian Network-based what-if analysis to see how the knowledge map between target node and related explanatory nodes could yield meaningful glimpse into what is going on in sentiments underlying the target dataset.

• Journal of KIISE:Computing Practices and Letters
• /
• v.9 no.3
• /
• pp.322-331
• /
• 2003

As the complexity of a 3D game is increased by various factors of the game scenario, it has a problem for controlling the interrelation of the game objects. Therefore, a game system has a necessity of the coordination of the responses of the game objects. Also, it is necessary to control the behaviors of animations of the game objects in terms of the game scenario. To produce realistic game simulations, a system has to include a structure for designing the interactions among the game objects. This paper presents a method that designs the dynamic control mechanism for the interaction of the game objects in the game scenario. For the method, we suggest a game agent system as a framework that is based on intelligent agents who can make decisions using specific rules. Game agent systems are used in order to manage environment data, to simulate the game objects, to control interactions among game objects, and to support visual authoring interface that ran define a various interrelations of the game objects. These techniques can process the autonomy level of the game objects and the associated collision avoidance method, etc. Also, it is possible to make the coherent decision-making ability of the game objects about a change of the scene. In this paper, the rule-based behavior control was designed to guide the simulation of the game objects. The rules are pre-defined by the user using visual interface for designing their interaction. The Agent State Decision Network, which is composed of the visual elements, is able to pass the information and infers the current state of the game objects. All of such methods can monitor and check a variation of motion state between game objects in real time. Finally, we present a validation of the control method together with a simple case-study example. In this paper, we design and implement the supervised classification systems for high resolution satellite images. The systems support various interfaces and statistical data of training samples so that we can select the most effective training data. In addition, the efficient extension of new classification algorithms and satellite image formats are applied easily through the modularized systems. The classifiers are considered the characteristics of spectral bands from the selected training data. They provide various supervised classification algorithms which include Parallelepiped, Minimum distance, Mahalanobis distance, Maximum likelihood and Fuzzy theory. We used IKONOS images for the input and verified the systems for the classification of high resolution satellite images.

• Journal of Intelligence and Information Systems
• /
• v.24 no.3
• /
• pp.21-44
• /
• 2018

In recent years, the rapid development of internet technology and the popularization of smart devices have resulted in massive amounts of text data. Those text data were produced and distributed through various media platforms such as World Wide Web, Internet news feeds, microblog, and social media. However, this enormous amount of easily obtained information is lack of organization. Therefore, this problem has raised the interest of many researchers in order to manage this huge amount of information. Further, this problem also required professionals that are capable of classifying relevant information and hence text classification is introduced. Text classification is a challenging task in modern data analysis, which it needs to assign a text document into one or more predefined categories or classes. In text classification field, there are different kinds of techniques available such as K-Nearest Neighbor, Naïve Bayes Algorithm, Support Vector Machine, Decision Tree, and Artificial Neural Network. However, while dealing with huge amount of text data, model performance and accuracy becomes a challenge. According to the type of words used in the corpus and type of features created for classification, the performance of a text classification model can be varied. Most of the attempts are been made based on proposing a new algorithm or modifying an existing algorithm. This kind of research can be said already reached their certain limitations for further improvements. In this study, aside from proposing a new algorithm or modifying the algorithm, we focus on searching a way to modify the use of data. It is widely known that classifier performance is influenced by the quality of training data upon which this classifier is built. The real world datasets in most of the time contain noise, or in other words noisy data, these can actually affect the decision made by the classifiers built from these data. In this study, we consider that the data from different domains, which is heterogeneous data might have the characteristics of noise which can be utilized in the classification process. In order to build the classifier, machine learning algorithm is performed based on the assumption that the characteristics of training data and target data are the same or very similar to each other. However, in the case of unstructured data such as text, the features are determined according to the vocabularies included in the document. If the viewpoints of the learning data and target data are different, the features may be appearing different between these two data. In this study, we attempt to improve the classification accuracy by strengthening the robustness of the document classifier through artificially injecting the noise into the process of constructing the document classifier. With data coming from various kind of sources, these data are likely formatted differently. These cause difficulties for traditional machine learning algorithms because they are not developed to recognize different type of data representation at one time and to put them together in same generalization. Therefore, in order to utilize heterogeneous data in the learning process of document classifier, we apply semi-supervised learning in our study. However, unlabeled data might have the possibility to degrade the performance of the document classifier. Therefore, we further proposed a method called Rule Selection-Based Ensemble Semi-Supervised Learning Algorithm (RSESLA) to select only the documents that contributing to the accuracy improvement of the classifier. RSESLA creates multiple views by manipulating the features using different types of classification models and different types of heterogeneous data. The most confident classification rules will be selected and applied for the final decision making. In this paper, three different types of real-world data sources were used, which are news, twitter and blogs.

• Journal of Intelligence and Information Systems
• /
• v.26 no.1
• /
• pp.23-45
• /
• 2020

Big data is creating in a wide variety of fields such as medical care, manufacturing, logistics, sales site, SNS, and the dataset characteristics are also diverse. In order to secure the competitiveness of companies, it is necessary to improve decision-making capacity using a classification algorithm. However, most of them do not have sufficient knowledge on what kind of classification algorithm is appropriate for a specific problem area. In other words, determining which classification algorithm is appropriate depending on the characteristics of the dataset was has been a task that required expertise and effort. This is because the relationship between the characteristics of datasets (called meta-features) and the performance of classification algorithms has not been fully understood. Moreover, there has been little research on meta-features reflecting the characteristics of multi-class. Therefore, the purpose of this study is to empirically analyze whether meta-features of multi-class datasets have a significant effect on the performance of classification algorithms. In this study, meta-features of multi-class datasets were identified into two factors, (the data structure and the data complexity,) and seven representative meta-features were selected. Among those, we included the Herfindahl-Hirschman Index (HHI), originally a market concentration measurement index, in the meta-features to replace IR(Imbalanced Ratio). Also, we developed a new index called Reverse ReLU Silhouette Score into the meta-feature set. Among the UCI Machine Learning Repository data, six representative datasets (Balance Scale, PageBlocks, Car Evaluation, User Knowledge-Modeling, Wine Quality(red), Contraceptive Method Choice) were selected. The class of each dataset was classified by using the classification algorithms (KNN, Logistic Regression, Nave Bayes, Random Forest, and SVM) selected in the study. For each dataset, we applied 10-fold cross validation method. 10% to 100% oversampling method is applied for each fold and meta-features of the dataset is measured. The meta-features selected are HHI, Number of Classes, Number of Features, Entropy, Reverse ReLU Silhouette Score, Nonlinearity of Linear Classifier, Hub Score. F1-score was selected as the dependent variable. As a result, the results of this study showed that the six meta-features including Reverse ReLU Silhouette Score and HHI proposed in this study have a significant effect on the classification performance. (1) The meta-features HHI proposed in this study was significant in the classification performance. (2) The number of variables has a significant effect on the classification performance, unlike the number of classes, but it has a positive effect. (3) The number of classes has a negative effect on the performance of classification. (4) Entropy has a significant effect on the performance of classification. (5) The Reverse ReLU Silhouette Score also significantly affects the classification performance at a significant level of 0.01. (6) The nonlinearity of linear classifiers has a significant negative effect on classification performance. In addition, the results of the analysis by the classification algorithms were also consistent. In the regression analysis by classification algorithm, Naïve Bayes algorithm does not have a significant effect on the number of variables unlike other classification algorithms. This study has two theoretical contributions: (1) two new meta-features (HHI, Reverse ReLU Silhouette score) was proved to be significant. (2) The effects of data characteristics on the performance of classification were investigated using meta-features. The practical contribution points (1) can be utilized in the development of classification algorithm recommendation system according to the characteristics of datasets. (2) Many data scientists are often testing by adjusting the parameters of the algorithm to find the optimal algorithm for the situation because the characteristics of the data are different. In this process, excessive waste of resources occurs due to hardware, cost, time, and manpower. This study is expected to be useful for machine learning, data mining researchers, practitioners, and machine learning-based system developers. The composition of this study consists of introduction, related research, research model, experiment, conclusion and discussion.

• Journal of Korea Society of Industrial Information Systems
• /
• v.12 no.4
• /
• pp.15-24
• /
• 2007

The recognition of postal address images is indispensable for the automatic sorting of postal envelopes. The process of the address image recognition is composed of three steps-address image preprocessing, character recognition, address interpretation. The extracted character images from the preprocessing step are forwarded to the character recognition step, in which multiple candidate characters with reliability scores are obtained for each character image extracted. aracters with reliability scores are obtained for each character image extracted. Utilizing those character candidates with scores, we obtain the final valid address for the input envelope image through the address interpretation step. The envelope sorting rate depends on the performance of all three steps, among which character recognition step could be said to be very important. The good character recognizer would be the one which could produce valid candidates with very reliable scores to help the address interpretation step go easy. In this paper, we propose the method of generating character candidates with reliable recognition scores. We utilize the existing MLP(multilayered perceptrons) neural network of the address recognition system in the current automatic postal envelope sorters, as the classifier for the each image from the preprocessing step. The MLP is well known to be one of the best classifiers in terms of processing speed and recognition rate. The false alarm problem, however, might be occurred in recognition results, which made the address interpretation hard. To make address interpretation easy and improve the envelope sorting rate, we propose promising methods to reestimate the recognition score (confidence) of the existing MLP classifier: the generation method of the statistical recognition properties of the classifier and the method of the combination of the MLP and the subspace classifier which roles as a reestimator of the confidence. To confirm the superiority of the proposed method, we have used the character images of the real postal envelopes from the sorters in the post office. The experimental results show that the proposed method produces high reliability in terms of error and rejection for individual characters and non-characters.

• Journal of Intelligence and Information Systems
• /
• v.20 no.3
• /
• pp.77-92
• /
• 2014

Recently, numerous documents including unstructured data and text have been created due to the rapid increase in the usage of social media and the Internet. Each document is usually provided with a specific category for the convenience of the users. In the past, the categorization was performed manually. However, in the case of manual categorization, not only can the accuracy of the categorization be not guaranteed but the categorization also requires a large amount of time and huge costs. Many studies have been conducted towards the automatic creation of categories to solve the limitations of manual categorization. Unfortunately, most of these methods cannot be applied to categorizing complex documents with multiple topics because the methods work by assuming that one document can be categorized into one category only. In order to overcome this limitation, some studies have attempted to categorize each document into multiple categories. However, they are also limited in that their learning process involves training using a multi-categorized document set. These methods therefore cannot be applied to multi-categorization of most documents unless multi-categorized training sets are provided. To overcome the limitation of the requirement of a multi-categorized training set by traditional multi-categorization algorithms, we propose a new methodology that can extend a category of a single-categorized document to multiple categorizes by analyzing relationships among categories, topics, and documents. First, we attempt to find the relationship between documents and topics by using the result of topic analysis for single-categorized documents. Second, we construct a correspondence table between topics and categories by investigating the relationship between them. Finally, we calculate the matching scores for each document to multiple categories. The results imply that a document can be classified into a certain category if and only if the matching score is higher than the predefined threshold. For example, we can classify a certain document into three categories that have larger matching scores than the predefined threshold. The main contribution of our study is that our methodology can improve the applicability of traditional multi-category classifiers by generating multi-categorized documents from single-categorized documents. Additionally, we propose a module for verifying the accuracy of the proposed methodology. For performance evaluation, we performed intensive experiments with news articles. News articles are clearly categorized based on the theme, whereas the use of vulgar language and slang is smaller than other usual text document. We collected news articles from July 2012 to June 2013. The articles exhibit large variations in terms of the number of types of categories. This is because readers have different levels of interest in each category. Additionally, the result is also attributed to the differences in the frequency of the events in each category. In order to minimize the distortion of the result from the number of articles in different categories, we extracted 3,000 articles equally from each of the eight categories. Therefore, the total number of articles used in our experiments was 24,000. The eight categories were "IT Science," "Economy," "Society," "Life and Culture," "World," "Sports," "Entertainment," and "Politics." By using the news articles that we collected, we calculated the document/category correspondence scores by utilizing topic/category and document/topics correspondence scores. The document/category correspondence score can be said to indicate the degree of correspondence of each document to a certain category. As a result, we could present two additional categories for each of the 23,089 documents. Precision, recall, and F-score were revealed to be 0.605, 0.629, and 0.617 respectively when only the top 1 predicted category was evaluated, whereas they were revealed to be 0.838, 0.290, and 0.431 when the top 1 - 3 predicted categories were considered. It was very interesting to find a large variation between the scores of the eight categories on precision, recall, and F-score.

• Journal of Intelligence and Information Systems
• /
• v.27 no.3
• /
• pp.157-173
• /
• 2021

With the development of information technology, the amount of available information increases daily. However, having access to so much information makes it difficult for users to easily find the information they seek. Users want a visualized system that reduces information retrieval and learning time, saving them from personally reading and judging all available information. As a result, recommendation systems are an increasingly important technologies that are essential to the business. Collaborative filtering is used in various fields with excellent performance because recommendations are made based on similar user interests and preferences. However, limitations do exist. Sparsity occurs when user-item preference information is insufficient, and is the main limitation of collaborative filtering. The evaluation value of the user item matrix may be distorted by the data depending on the popularity of the product, or there may be new users who have not yet evaluated the value. The lack of historical data to identify consumer preferences is referred to as data sparsity, and various methods have been studied to address these problems. However, most attempts to solve the sparsity problem are not optimal because they can only be applied when additional data such as users' personal information, social networks, or characteristics of items are included. Another problem is that real-world score data are mostly biased to high scores, resulting in severe imbalances. One cause of this imbalance distribution is the purchasing bias, in which only users with high product ratings purchase products, so those with low ratings are less likely to purchase products and thus do not leave negative product reviews. Due to these characteristics, unlike most users' actual preferences, reviews by users who purchase products are more likely to be positive. Therefore, the actual rating data is over-learned in many classes with high incidence due to its biased characteristics, distorting the market. Applying collaborative filtering to these imbalanced data leads to poor recommendation performance due to excessive learning of biased classes. Traditional oversampling techniques to address this problem are likely to cause overfitting because they repeat the same data, which acts as noise in learning, reducing recommendation performance. In addition, pre-processing methods for most existing data imbalance problems are designed and used for binary classes. Binary class imbalance techniques are difficult to apply to multi-class problems because they cannot model multi-class problems, such as objects at cross-class boundaries or objects overlapping multiple classes. To solve this problem, research has been conducted to convert and apply multi-class problems to binary class problems. However, simplification of multi-class problems can cause potential classification errors when combined with the results of classifiers learned from other sub-problems, resulting in loss of important information about relationships beyond the selected items. Therefore, it is necessary to develop more effective methods to address multi-class imbalance problems. We propose a collaborative filtering model using CGAN to generate realistic virtual data to populate the empty user-item matrix. Conditional vector y identify distributions for minority classes and generate data reflecting their characteristics. Collaborative filtering then maximizes the performance of the recommendation system via hyperparameter tuning. This process should improve the accuracy of the model by addressing the sparsity problem of collaborative filtering implementations while mitigating data imbalances arising from real data. Our model has superior recommendation performance over existing oversampling techniques and existing real-world data with data sparsity. SMOTE, Borderline SMOTE, SVM-SMOTE, ADASYN, and GAN were used as comparative models and we demonstrate the highest prediction accuracy on the RMSE and MAE evaluation scales. Through this study, oversampling based on deep learning will be able to further refine the performance of recommendation systems using actual data and be used to build business recommendation systems.

• Journal of Intelligence and Information Systems
• /
• v.17 no.1
• /
• pp.53-69
• /
• 2011

Vision and voice-based technologies are commonly utilized for human-robot interaction. But it is widely recognized that the performance of vision and voice-based interaction systems is deteriorated by a large margin in the real-world situations due to environmental and user variances. Human users need to be very cooperative to get reasonable performance, which significantly limits the usability of the vision and voice-based human-robot interaction technologies. As a result, touch screens are still the major medium of human-robot interaction for the real-world applications. To empower the usability of robots for various services, alternative interaction technologies should be developed to complement the problems of vision and voice-based technologies. In this paper, we propose the use of accelerometer-based gesture interface as one of the alternative technologies, because accelerometers are effective in detecting the movements of human body, while their performance is not limited by environmental contexts such as lighting conditions or camera's field-of-view. Moreover, accelerometers are widely available nowadays in many mobile devices. We tackle the problem of classifying acceleration signal patterns of 26 English alphabets, which is one of the essential repertoires for the realization of education services based on robots. Recognizing 26 English handwriting patterns based on accelerometers is a very difficult task to take over because of its large scale of pattern classes and the complexity of each pattern. The most difficult problem that has been undertaken which is similar to our problem was recognizing acceleration signal patterns of 10 handwritten digits. Most previous studies dealt with pattern sets of 8~10 simple and easily distinguishable gestures that are useful for controlling home appliances, computer applications, robots etc. Good features are essential for the success of pattern recognition. To promote the discriminative power upon complex English alphabet patterns, we extracted 'motion trajectories' out of input acceleration signal and used them as the main feature. Investigative experiments showed that classifiers based on trajectory performed 3%~5% better than those with raw features e.g. acceleration signal itself or statistical figures. To minimize the distortion of trajectories, we applied a simple but effective set of smoothing filters and band-pass filters. It is well known that acceleration patterns for the same gesture is very different among different performers. To tackle the problem, online incremental learning is applied for our system to make it adaptive to the users' distinctive motion properties. Our system is based on instance-based learning (IBL) where each training sample is memorized as a reference pattern. Brute-force incremental learning in IBL continuously accumulates reference patterns, which is a problem because it not only slows down the classification but also downgrades the recall performance. Regarding the latter phenomenon, we observed a tendency that as the number of reference patterns grows, some reference patterns contribute more to the false positive classification. Thus, we devised an algorithm for optimizing the reference pattern set based on the positive and negative contribution of each reference pattern. The algorithm is performed periodically to remove reference patterns that have a very low positive contribution or a high negative contribution. Experiments were performed on 6500 gesture patterns collected from 50 adults of 30~50 years old. Each alphabet was performed 5 times per participant using $Nintendo{(R)}$ $Wii^{TM}$ remote. Acceleration signal was sampled in 100hz on 3 axes. Mean recall rate for all the alphabets was 95.48%. Some alphabets recorded very low recall rate and exhibited very high pairwise confusion rate. Major confusion pairs are D(88%) and P(74%), I(81%) and U(75%), N(88%) and W(100%). Though W was recalled perfectly, it contributed much to the false positive classification of N. By comparison with major previous results from VTT (96% for 8 control gestures), CMU (97% for 10 control gestures) and Samsung Electronics(97% for 10 digits and a control gesture), we could find that the performance of our system is superior regarding the number of pattern classes and the complexity of patterns. Using our gesture interaction system, we conducted 2 case studies of robot-based edutainment services. The services were implemented on various robot platforms and mobile devices including $iPhone^{TM}$. The participating children exhibited improved concentration and active reaction on the service with our gesture interface. To prove the effectiveness of our gesture interface, a test was taken by the children after experiencing an English teaching service. The test result showed that those who played with the gesture interface-based robot content marked 10% better score than those with conventional teaching. We conclude that the accelerometer-based gesture interface is a promising technology for flourishing real-world robot-based services and content by complementing the limits of today's conventional interfaces e.g. touch screen, vision and voice.

• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.167-181
• /
• 2018

Over the past decade, deep learning has been in spotlight among various machine learning algorithms. In particular, CNN(Convolutional Neural Network), which is known as the effective solution for recognizing and classifying images or voices, has been popularly applied to classification and prediction problems. In this study, we investigate the way to apply CNN in business problem solving. Specifically, this study propose to apply CNN to stock market prediction, one of the most challenging tasks in the machine learning research. As mentioned, CNN has strength in interpreting images. Thus, the model proposed in this study adopts CNN as the binary classifier that predicts stock market direction (upward or downward) by using time series graphs as its inputs. That is, our proposal is to build a machine learning algorithm that mimics an experts called 'technical analysts' who examine the graph of past price movement, and predict future financial price movements. Our proposed model named 'CNN-FG(Convolutional Neural Network using Fluctuation Graph)' consists of five steps. In the first step, it divides the dataset into the intervals of 5 days. And then, it creates time series graphs for the divided dataset in step 2. The size of the image in which the graph is drawn is $40(pixels){\times}40(pixels)$, and the graph of each independent variable was drawn using different colors. In step 3, the model converts the images into the matrices. Each image is converted into the combination of three matrices in order to express the value of the color using R(red), G(green), and B(blue) scale. In the next step, it splits the dataset of the graph images into training and validation datasets. We used 80% of the total dataset as the training dataset, and the remaining 20% as the validation dataset. And then, CNN classifiers are trained using the images of training dataset in the final step. Regarding the parameters of CNN-FG, we adopted two convolution filters ($5{\times}5{\times}6$ and $5{\times}5{\times}9$) in the convolution layer. In the pooling layer, $2{\times}2$ max pooling filter was used. The numbers of the nodes in two hidden layers were set to, respectively, 900 and 32, and the number of the nodes in the output layer was set to 2(one is for the prediction of upward trend, and the other one is for downward trend). Activation functions for the convolution layer and the hidden layer were set to ReLU(Rectified Linear Unit), and one for the output layer set to Softmax function. To validate our model - CNN-FG, we applied it to the prediction of KOSPI200 for 2,026 days in eight years (from 2009 to 2016). To match the proportions of the two groups in the independent variable (i.e. tomorrow's stock market movement), we selected 1,950 samples by applying random sampling. Finally, we built the training dataset using 80% of the total dataset (1,560 samples), and the validation dataset using 20% (390 samples). The dependent variables of the experimental dataset included twelve technical indicators popularly been used in the previous studies. They include Stochastic %K, Stochastic %D, Momentum, ROC(rate of change), LW %R(Larry William's %R), A/D oscillator(accumulation/distribution oscillator), OSCP(price oscillator), CCI(commodity channel index), and so on. To confirm the superiority of CNN-FG, we compared its prediction accuracy with the ones of other classification models. Experimental results showed that CNN-FG outperforms LOGIT(logistic regression), ANN(artificial neural network), and SVM(support vector machine) with the statistical significance. These empirical results imply that converting time series business data into graphs and building CNN-based classification models using these graphs can be effective from the perspective of prediction accuracy. Thus, this paper sheds a light on how to apply deep learning techniques to the domain of business problem solving.

• Journal of Intelligence and Information Systems
• /
• v.19 no.4
• /
• pp.123-132
• /
• 2013

As the smartphones are equipped with various sensors such as the accelerometer, GPS, gravity sensor, gyros, ambient light sensor, proximity sensor, and so on, there have been many research works on making use of these sensors to create valuable applications. Human activity recognition is one such application that is motivated by various welfare applications such as the support for the elderly, measurement of calorie consumption, analysis of lifestyles, analysis of exercise patterns, and so on. One of the challenges faced when using the smartphone sensors for activity recognition is that the number of sensors used should be minimized to save the battery power. When the number of sensors used are restricted, it is difficult to realize a highly accurate activity recognizer or a classifier because it is hard to distinguish between subtly different activities relying on only limited information. The difficulty gets especially severe when the number of different activity classes to be distinguished is very large. In this paper, we show that a fairly accurate classifier can be built that can distinguish ten different activities by using only a single sensor data, i.e., the smartphone accelerometer data. The approach that we take to dealing with this ten-class problem is to use the ensemble of nested dichotomy (END) method that transforms a multi-class problem into multiple two-class problems. END builds a committee of binary classifiers in a nested fashion using a binary tree. At the root of the binary tree, the set of all the classes are split into two subsets of classes by using a binary classifier. At a child node of the tree, a subset of classes is again split into two smaller subsets by using another binary classifier. Continuing in this way, we can obtain a binary tree where each leaf node contains a single class. This binary tree can be viewed as a nested dichotomy that can make multi-class predictions. Depending on how a set of classes are split into two subsets at each node, the final tree that we obtain can be different. Since there can be some classes that are correlated, a particular tree may perform better than the others. However, we can hardly identify the best tree without deep domain knowledge. The END method copes with this problem by building multiple dichotomy trees randomly during learning, and then combining the predictions made by each tree during classification. The END method is generally known to perform well even when the base learner is unable to model complex decision boundaries As the base classifier at each node of the dichotomy, we have used another ensemble classifier called the random forest. A random forest is built by repeatedly generating a decision tree each time with a different random subset of features using a bootstrap sample. By combining bagging with random feature subset selection, a random forest enjoys the advantage of having more diverse ensemble members than a simple bagging. As an overall result, our ensemble of nested dichotomy can actually be seen as a committee of committees of decision trees that can deal with a multi-class problem with high accuracy. The ten classes of activities that we distinguish in this paper are 'Sitting', 'Standing', 'Walking', 'Running', 'Walking Uphill', 'Walking Downhill', 'Running Uphill', 'Running Downhill', 'Falling', and 'Hobbling'. The features used for classifying these activities include not only the magnitude of acceleration vector at each time point but also the maximum, the minimum, and the standard deviation of vector magnitude within a time window of the last 2 seconds, etc. For experiments to compare the performance of END with those of other methods, the accelerometer data has been collected at every 0.1 second for 2 minutes for each activity from 5 volunteers. Among these 5,900 ($=5{\times}(60{\times}2-2)/0.1$) data collected for each activity (the data for the first 2 seconds are trashed because they do not have time window data), 4,700 have been used for training and the rest for testing. Although 'Walking Uphill' is often confused with some other similar activities, END has been found to classify all of the ten activities with a fairly high accuracy of 98.4%. On the other hand, the accuracies achieved by a decision tree, a k-nearest neighbor, and a one-versus-rest support vector machine have been observed as 97.6%, 96.5%, and 97.6%, respectively.