Hayato Ohwada

Identifying Driver’s Cognitive Load Using Inductive Logic Programming

This paper uses inductive logic programming (ILP) to identify a driver’s cognitive state in real driving situations to determine whether a driver will be ready to select a suitable operation and recommended service in the next generation car navigation systems. We measure the driver’s eye movement and collect various data such as braking, acceleration and steering angles that are qualitatively interpreted and represented as background knowledge. A set of data about the driver’s degree of tension or relaxation regarded as a training set is obtained from the driver’s mental load based on resource-limited cognitive process analysis. Given such information, our ILP system has successfully produced logic rules that are qualitatively understandable for rule verification and are actively employed for user-oriented interface design. Realistic experiments were conducted to demonstrate the learning performance of this approach. Reasonable accuracy was achieved for an appropriate service providing safe driving.

Shill bidder detection for online auctions

Recently, the online auction has become a popular Internet service. Since the service has been expanded rapidly, security risks in the system remain. Fundamental measures are still required. This paper proposes a method for detecting shill bidders in online auctions. It first detects outliers with a oneclass SVM. It then transforms the results into a decision tree using C4.5. The experiment results demonstrate that we can use the resulting rules to classify shill bidders.

Intercommunicating car navigation system with dynamic route finding

We propose a dynamic route finding method based on time-constrained search (TCS) that finds a provably optimal solution within a specified time. Our car navigation system collects information through the local communication with other systems using wireless LAN, and it can generate a new route to avoid the traffic jam by setting the time limit as the time for an automobile to go to the road the traffic jam happens. Our experiments with a digital map shows that the estimation of the finishing time of the route finding is quite accurate, and optimal solutions are produced by making full use of the permissible search time.

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.

Classifying Cognitive Load and Driving Situation with Machine Learning

 Abstract—This paper classifies a drivers cognitive state in real driving situations to improve the in-vehicle information service that judges a users cognitive load and driving situation. We measure the drivers eye movement and collect driving sensor data such as braking, acceleration, and steering angles that are used to classify the drivers state. A set of data about the drivers degree of cognitive load, regarded as a training set, is obtained from steering operation and task cognition. Given such information, we use a machine-learning method to classify the drivers cognitive load. We achieved reasonable accuracy in certain driving situations in which the driver moves abnormally for an appropriate service supporting safe driving. surrounding situation, the range of eye movement may increase because the drivers cognitive load has increased. This means that a high cognitive load and a change of eye movement are related. Eye movement is used in the field of physiological psychology for clarifying control (1). It is directly related to perception and can be considered an indication of mental load. Driving a car requires prediction of the surrounding environment and is influenced by the situation. Therefore, a users cognitive load can be classified using these features. To do this, we measure the drivers eye movement and gather driving data such as accelerator use, braking, and steering. This paper takes a machine-learning approach to the above cognitive state identification problem in a realistic car-driving task. We set up cognitive loads according to the steering-entropy method (2) and a definition of the task cognition situation. This paper is organized as follows. Section II presents related works. Section III defines cognitive load. Section IV describes our classification model for cognitive load. Section V presents a performance evaluation. The final section provides conclusions.

Utilizing finger movement data to cluster patients with everyday action impairment

Difficulty in performing the activities of daily living is a key clinical feature of early cognitive decline in older adults and has also been associated with the early stage of dementia in mild cognitive impairment (MCI). As the number of individuals with dementia and the development of technology rise, an immersive virtual environment or virtual reality has been used in therapy for memory problems in the area of MCI. This study evaluated the use of finger movement data obtained from the virtual-reality-based application and its ability to cluster patients with everyday action impairment. Here, as a pilot study, nine healthy adults completed lunch box packing as an everyday action task in the designated virtual reality called the Virtual Kitchen (VK), equipped with a leap motion controller to record their finger movement. We converted the finger movements to acceleration data and then employed a time series clustering algorithm to create several clusters based on the data set. In addition, to comprehensively review the clustering result, we assessed performance-based measures for the experiment using the Naturalistic Action Test (NAT). The final results indicate that the clusters formed by using the acceleration data seem reasonably analogous to the performance measures (i.e., the type and number of errors that occurred).

Applicability of machine-learning techniques in predicting customer defection

Machine learning is an established method of predicting customer defection from a contractual business. However, no systematic comparison or evaluation of the different machine-learning techniques has been performed. In this study, we provide a comprehensive comparison of different machine-learning techniques with three different data sets of a software company to predict customer defection. The evaluation criteria of the techniques are understandability of the model, convenience of using the model, time efficiency in running the learning model, and performance of predicting customer defection.

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.

Proximity Scoring Using Sentence-Based Inverted Index for Practical Full-Text Search

We propose a search method that uses sentence-based inverted indexes to achieve high accuracy at practical speeds. The proposed method well supports the vast majority of queries entered on the web; these queries contain single words, multiple words for proximity searches, and semantically direct phrases. The existing approach, the inverted index which holds word-level position data is not efficient, because the size of index becomes extremely large. Our solution is to drop the word position data and index only the existence of each word in each sentence. We incorporate the sentence-based inverted index into a commercial search engine and evaluate it using both Japanese and English standard IR corpuses. The experiment shows that our method offers high accuracy, while index size and search processing time are greatly reduced.

Using Inductive Logic Programming to Learn Rules that Identify Glaucomatous Eyes

This chapter examines the applicability and performance of Inductive Logic Programming (ILP) in learning classification rules for a medical domain. The domain is glaucoma diagnosis where ocular fundus images are used to identify glaucomatous eyes. An ILP system called GKS was developed, not only to deal with low-level measurement data such as images but also produce diagnostic rules that are readable and comprehensive for interactions with medical experts. Since such rules are directly used as diagnostic rules, the present method provides automatic construction of a knowledge base from an expert’s accumulated diagnostic experience. A variety of experiments are conducted to clarify the performance of classification based on the induced rules. The resulting performance is comparable with human-level classification. This indicates that an ILP-based method can be used as a highly-valuable medical decision tool.