Katsutoshi Kanamori

Customer Lifetime Value and Defection Possibility Prediction Model Using Machine Learning: An Application to a Cloud-Based Software Company

This paper proposes an estimation of Customer Lifetime Value (CLV) for a cloud-based software company by using machine learning techniques. The purpose of this study is twofold. We classify the customers of one cloud-based software company by using two classifications methods: C4.5 and a support vector machine (SVM). We use machine learning primarily to estimate the frequency distribution of the customer defection possibility. The result shows that both the C4.5 and SVM classifications perform well, and by obtaining frequency distributions of the defection possibility, we can predict the number of customers defecting and the number of customers retained.

A Route Search System in Consideration of the Reservation Service in Amusement Parks for Smart Phone

Many amusement parks adopt a reservation service(e.g. Fast pass at Disneyland) , that effectively reduce the waiting time for visitors. Even if visitors do use the reservation service, the traveling time may be long, depending on the order in which users visit the attractions. We think that people need a new route search algorithm to enhance the reservation service. Therefore, we have developed a new algorithm employingstructured programming. We constructed the system to be executed on a smart phone by using constraint logic programming and Java.

Trust-Aware Recommender System Incorporating Review Contents

Personalized recommendation systems can help people find things that interest them and are widely used in developing the Internet or e-commerce. Collaborative filtering (CF) seems to be the most popular technique in recommender systems. However, CF is weak in the process of finding similar users. To resolve these problems, trust-aware recommender systems (TaRSs) have been developed in recent years. In this study, we propose a new approach that incorporates the content of reviews in a TaRS. In addition, we use a new dataset that is collected from the Yahoo!Movie website, whereas traditional research has used Epinions or Movielens. Finally, we evaluate the experiment results using precision and coverage.

Flexible route planning for amusement parks navigation

This paper describes flexible route planning for amusement parks (e.g. Disneyland) navigation. Unlike conventional shortest path finding, we provide several types of algorithms that consider waiting time estimation in real time, exploit the reservation facility of an attraction such as Fast Pass in Disneyland, and balance a series of enjoyment types such as excitement or relaxation. These features extend Dikstras shortest path algorithm to be more flexible and dynamical. We have developed a navigation tool as a Web application in which users select their interesting attractions and then the application suggests reasonable and enjoyable routes. The experiment was conducted to show the performance of this application focusing on well-known attractions in Tokyo Disneyland.

Gene Selection from Microarray Data for Alzheimer's Disease Using Random Forest

A significant amount of microarray gene expression data is available on the Internet, and researchers are allowed to analyze such data freely. However, microarray data includes thousands of genes, and analysis using conventional techniques is too difficult. Therefore, selecting informative gene(s) from high-dimensional data is very important. In this study, the authors propose a gene selection method using random forest as a machine learning technique. They applied this method to microarray data on Alzheimer’s disease and conducted an experiment to rank genes. The authors’ results indicated some genes that have been investigated for their relevance to Alzheimer’s disease, proving that their proposed cognitive method was successful in finding disease-related genes using microarray data.

Extracting time-oriented relationships of nutrients to losing body fat mass using inductive logic programming

This study was performed to extract rules for reducing body fat mass so as to prevent lifestyle-related diseases. Lifestyle-related diseases have been increasing in Japan, even among younger people. Body fat mass is related to lifestyle-related diseases. Hence, finding rules for reducing body fat mass is very meaningful. We obtained lifestyle time-series data on five male subjects who are in their 20s and not obese. The data includes the amount of body fat mass of each subject and a variety of features such as sleep, exercise, and nutrient intake. We used Inductive Logic Programming (ILP) to apply this data because ILP can more flexibly learn rules than other machine-learning methods. As a result of applying the data to ILP, our ILP system successfully extracted rules of time-oriented relationships of nutrients to decrease body fat mass based on limited data. Intake of various nutrients one day and two days prior was effective in reducing body fat mass. Moreover, we determined that nutrients related to losing body fat mass include vitamin B2, pantothenic acid, fat, vitamin B1, and biotin.

Finding a disease-related gene from microarray data using random forest

Numerous databases of DNA-microarrays are now widely available on the internet. Recently, there has been increasing interest in the analysis of microarray data using machine-learning techniques due to the amount of data, which is too massive for researchers to analyze using conventional techniques. In this study, we propose a method of finding a disease-related gene from microarray data using random forest, a machine-learning technique. More specifically, we focused on Alzheimers disease and used microarray data related to Alzheimers disease in the experiments. In the result, we found some genes that are believed to be related to Alzheimers disease. Some genes discovered in the result have been investigated for their relevance to Alzheimers disease, and this proves that our proposed methodology was successful in finding disease-related genes using microarray data. In addition, the proposed methodology is useful in providing new knowledge for biologists, medical scientists, and cognitive computing researchers since there is no previous work on genes that focused on finding a disease-related gene for Alzheimers disease.

Combining Feature Selection with Decision Tree Criteria and Neural Network for Corporate Value Classification

This study aims to classify corporate values among Japanese companies based on their corporate social responsibility (CSR) performances. Since there are many attributes in CSR, feature selection with decision tree criteria is used to select the attributes that can classify corporate values. The feature selection found that 41 % of 37 total attributes, or only 15 attributes, are needed to classify corporate values. The accuracy of building the tree used to find the 15 attributes is low. To increase the accuracy, the attributes are trained in a neural network. The accuracy of the decision tree is 0.7, and the accuracy of the neural for training the 15 attributes increased to 0.75. To sum up, this study found, companies with higher corporate values seek to enhance their CSR activities or to empower secondary stakeholders. In contrast, companies with low corporate values still focus their CSR activities on primary stakeholders.

Implementing Majority Voting Rule to Classify Corporate Value Based on Environmental Efforts

The Japanese understanding of corporate social responsibility (CSR) is linked with the country’s history of industrial pollution. As a result, the top area Japanese companies are addressing is the environment. This study aims to classify the corporate value of Japanese companies calculated by the Ohlson model based on environmental efforts using several classification techniques. The corporate value is divided into high, medium, and low. Since the classification leads to imbalanced classes, five classification techniques (Gradient Boosting, Decision Tree, Support Vector Machine (SVM), and K-Nearest Neighbor (KNN)) were chosen to deal with this problem. KNN, with the lowest accuracy (0.68), was found predict smaller classes better than the others. To improve its accuracy, a majority voting rule is implemented in this study. In the voting rule, three classifiers (KNN, Random Forest, and Decision Tree) are combined. The accuracy for the combination of the three classifiers is 0.71. However, this study found that the impact on biodiversity is the most important variable among Japanese companies. This indicates that recent efforts to differentiate corporate value among Japanese companies based on environmental efforts arises from their understanding of the impact of business activities on biodiversity.

Various Classifiers to Investigate the Relationship Between CSR Activities and Corporate Value

The relationship between corporate social responsibility (CSR) and financial performance is complex and nuanced. Many studies have reported positive, negative, and neutral impacts of CSR on financial performance. This inconsistency is due to differences in methodologies, approaches, and selection of variables. Rather than focusing on specific variables, the present study aims to classify as many variables as possible in CSR if they contribute to shaping corporate value. In this study, we calculate corporate value using the Ohlson model based on income, since many previous studies focus on only a market-based approach. We chose some common classifiers that were appropriate for the nature of our data. After evaluating the performance of each classifier, we found that the Decision Tree is the best classifier to analyze the relationship between CSR activities and corporate value. Based on the tree, companies with high or medium corporate values seek to enhance their CSR activities or to empower secondary stakeholders (e.g., communities, societies), as indicated by cooperation with NPO/NGO. In contrast, companies with low corporate values still focus their CSR activities on primary stakeholders (e.g., customers, employees).