This paper focuses on the on-line learning method called Adaptive Regularization of Weight Vectors (AROW). AROW has two advantages compared to other on-line learning methods. First, it is robust to label noise. Second, we can obtain stable models by taking confidence intervals of parameters into consideration. However, AROW cannot perform feature selection. This paper proposes a novel method by combining AROW with lasso. We also employ the coordinate descent algorithm to estimate parameters, which enables us to speed up our algorithm. We confirm the effectiveness of our proposed method by some numerical experiments.
An alternating renewal process is used as modeling of various phenomena, such as two state system machine that periodically fails and repair. The reliability engineering the availability of system is of interest. That is defined by a probability that a system is on at any given time t. On-times and off-times appear alternately. We can sometimes obtain only partial observations through the limited observation windows. Only in this window we can observe the occurrence the specified events. In this article we propose a new estimation procedure for an availability parameter based on the window-censored observations. We derive the likelihood function corresponding to the window-censoring mechanism. In the case of Weibull distribution we show the usefulness of the proposed method. When the shape parameter is equal to one which is corresponding, the simple estimator (the ratio of total on-times to window size) that is easy calculated, is can be used practically. ML estimation has a good performance in other shape parameter values.
In this paper, we introduce a proccess of the data analysis for the POS data of MUJI provided by Ryohin Keikaku Co.,Ltd. in Data Analysis Competition 2014 hosted by Joint Association Study Group of Management Science. We found that many items which has been used to be considerd that women love were also bought heavily by men. So we assumed that the existence of such kind of customer segment named "Jyoshiryoku Danshi". We applied the hiarchical clustering to the customer's purchase count data collected with each item categories and the cluster which has the properties of "Jyoshiryoku Danshi" could be identified. At the same time, the relationship between the tweets from the official Twitter account of MUJI and the sales of the MUJI products was investigated. Based on these analyses, we proposed the effective marketing to "Jyoshiryoku Danshi" customer segment using Twitter.
Many stores-style exists so that it is said a characteristic an item to treat at each store as for the no mark quality goods to handle many items. However, in this study, I performed cluster analysis and investigated validity of the store-style of the no mark quality goods to confirm it whether those stores-style did not have to think about one or other stores-style made use of. In addition, using association rule analysis, I investigated whether even the cluster was different in the simultaneous purchasing of the product which I purchased. It is thought that this difference is useful for the sales strategy in the cluster.
We had developed a visualization software product with an animation for display the time series data set. We had proposed visualization of the amount of change using the software. It is differences of the aggregated values in groups from the time to the next time. We had analysis the data set that was provided for the date competition organized by the Joint Association Study Group of Management Science. Proposed software draw aggregated values which are divided by time, in each group with the extended parallel coordinated plot and the histogram. These graphics are shown by an animation in order of time. We had implemented it by the Java language and MySQL.
Customer each has buying pattern. In this study, We aimed to classification of customer buying pattern. If it is possible to classify customers by any criteria, we can suggest a different recommendation for each of the clusters. We used Self Organizing Maps (SOM) in order to classify of the customers. Classification by using SOM, there is no need to determine in advance the number of clusters. Also, it is applicable to big data, such as the POS data. In addition, we analyze about charactaristics of customer and group store and visualize the result of classification in the map.
We had proposed a visualization software product for the purchasing information using the three-dimensional computer graphics (3DCG) technology and explain results of proposed software. It can draw four variables in the three dimensional space. Three variables of them divide the data set into some groups. These groups are displayed as cubes. Aggregated values which are calculated by one variable in each group, are shown with a color of the cube. A large value is a dark color and a small value is a light color. We can also take aspects out of cubes to see their detail or compare them. They are displayed like a heat-map. We had implemented the software using the Java language and MySQL. We had visualized the data set that was provided for the date competition organized by the Joint Association Study Group of Management Science, and explained results of trend of the purchasing information.