Haruaki Yamazaki

Automatic sleep stage scoring based on waveform recognition method and decision-tree learning

There have been many trials in which the waveform recognition method, which is intended to replace expert observation by computer processing, has been used to extract the features of biological signals during sleep, and to automatically score sleep stages based on feature parameters. This paper proposes a waveform recognition method which extracts the feature parameters based on the characteristics of the biological signal during sleep, and a method of automatic sleep stage scoring by decision-tree learning, which is currently considered to be one of the most successful machine learning methods in practice. As the first step in the method, the features corresponding to the state of the EEG, the EOG, and the EMG during sleep are compared to the features of characteristic waves such as the α-wave, δ-wave, sleep spindle, K-complex, and REM, and the feature parameters needed in order to judge the sleep stage are extracted. Using canonical discriminant analysis and discretization method RWS based on the random walk, the feature parameters are converted to a small number of discrete variables. Based on training instances, obtained by the bootstrap method, a set of multiple small decision trees (a committee) is formed, and the sleep stage is scored by majority decision in the classification results. The method is applied to the PSG chart digital data provided by the Japan Sleep Society, and the performance of the system is evaluated experimentally. It is verified that the proposed method is promising as a method of automatic sleep stage scoring with high accuracy, requiring little expenditure of time in learning and classification.

Automated sleep stage scoring by decision tree learning

We describe a waveform recognition method that extracts characteristic parameters from waveforms and a method of automated sleep stage scoring using decision tree learning that is in practice regarded as one of the most successful machine learning methods. In our method, first characteristics of EEG, EOG and EMG are compared with characteristic features of alpha waves, delta waves, sleep spindles, K-complexes and REMs. Then, several parameters that are necessary for sleep stage scoring are extracted. We transform these extracted parameters into a few discrete variables using canonical discriminant analysis and the discretization method based on a random walk, and then a committee that consists of several small decision trees is formed from a small number of training instances. Furthermore final sleep stages are decided by a majority decision of the committee. Our method was applied to the digitized PSG chart data, provided by the Japan Society of Sleep Research and we carried out an evaluation experiment. The experiment indicated that our method can quickly execute learning and classification and precisely score sleep stages.

Multidirectional associative memory with a hidden layer

MAM (Multidirectional Associative Memory) is an extended BAM (Bidirectional Associative Memory), and an associative memory model which can deal with multiple associations. If the training set has common terms, the conventional MAM often recalls the convolutional patterns. IMAM (Improved Multidirectional Associative Memory) can store them, but the structure is complex and the storage capacity is extremely small because it must use correlation matrix. In this paper, we propose a MAM with a hidden layer and its learning method. The structure is as simple as MAM and can store the training set which includes common terms. By computer simulation, we show the storage capacity is far larger than correlation learning and it is robust against noise.

Extracting Thai compound nouns for paragraph extraction in Thai text

In this paper, we propose a method for Thai text summarization by paragraph extraction based on the extracted Thai compound nouns and term weighting method in terms of term frequency inverse document frequency (TF/spl middot/IDF). According to the highly frequent and highly productive of Thai compound nouns in Thai text, this property shows that Thai compound nouns play the important role in summarization. The morphological analysis is used to determine Thai compound nouns and all paragraphs are ranked by summation of term weighting score. The cosine similarity between each paragraph is calculated in order to select the important paragraphs among all paragraphs. The result shows that 0.469 F-score for 45% summary of our proposed method yield the most effective approach among all experiments.

RWS (random walk splitting): a random walk based discretization of continuous attributes

The discretization of continuous attributes in a given training set is an important issue, which significantly affects the performance of decision trees. This paper proposes a method to discretize the continuous attributes based on a random walk modeled statistical test. In this method, the algorithm tries to find the point which divides the training set T into two groups T1 and T2 such that T = T1 ∪ T2 with possibly many instances from a majority class included in T1. In other words, the algorithm detects the splitting point, which gives the maximum discrepancy between the two empirical distributions, the majority class and the rest. The algorithm recursively executes this procedure until some statistical criterion is satisfied. Further, we report the effectiveness of the algorithm over ChiMerge and MDLPC based on an experiment with UCI repository.