American football is a major commercial sport in the United States. There are many complicated situations in football, and it is considered a sport where statistical analysis and play prediction are difficult. In the United States, football analysis has been actively conducted and utilized in actual games. However, in Japan, merely a few teams have used statistical analysis in actual games.
To contribute to the Japanese football industry, we performed several analyses using statistical methods that enable quantitative interpretation by quantifying and visualizing football. Consequently, this research has enabled the analysis, quantification, and visualization of various elements of football. For example, the expected score value is calculated for each football situation, thereby facilitating a quantitative comparison of the characteristics of each team. Furthermore, among the prediction methods using machine learning, we can grasp the information that is compatible with the prediction of football plays that are run and passed, as well as the information necessary to construct a highly accurate classification rule.
Supersaturated designs provide the means to reduce a large number of factors down to a few active factors that influence the response of experimental results. Various data analysis methods for supersaturated designs are proposed to address the difficulty of factor estimation. However, few proposals have been made regarding which supersaturated design and data analysis method should be used to most accurately identify the active factors in computer experiments. This paper describes a series of numerical evaluation conducted with the aim of determining which combination of supersaturated design and data analysis method works better in accomplishing this task. Several models for computer experiments are examined using various combinations of existing supersaturated designs and analysis methods. We use 12×22,12×66.24×46,24×69,48×93 supersaturated designs, with Forward Stepwise selection, Model-Averaging, LASSO, and the Dantzig-Selector as the analysis methods. Numerical evaluation indicates that there is no single combination that always produces good results. Rather, it is found that the best combination depends on the conditions including design size, magnitude of active factors and so forth.It would appear that it is better to select the combination when the number of experiments is small or when the number of columns is large. In addition, the sign of the effect can influence the results for some supersaturated designs. To manage this problem, we propose a supersaturated design that gives relatively stable results regardless of the sign of the effect.
In professional baseball games, the timing of changing the starting pitcher is an important factor in winning a game. The exploitation of various sources of data was recently enabled. Thus, it is desirable to develop an effective analytical model for team management that learns from past data. In this study, we propose a statistical model to estimate the expected runs for each inning and support decision-making on changing starting pitchers. Specifically, we developed a Markov chain model with a transition probability matrix for states defined by the combination of out counts and occupation of the bases. Note that the transition probability between the pitcher and batter can vary depending on the match. However, even if the transition probabilities are estimated for each match, the number of combinations becomes very high, and the estimation accuracy is reduced. Therefore, we introduce a latent-class model to estimate the transition probabilities while grouping the combinations into a small number of latent variables. Using the proposed model, it is possible to estimate the expected runs accurately by using the transition probabilities estimated for each latent class. To verify the effectiveness of the proposed model, we conducted experiments using actual Japanese professional baseball data.
To assure quality of healthcare, it is important to develop the competence of healthcare professionals. As information technology introduction to clinical sites, the data acquisition in healthcare operation can be contributed to effective training. At present, however, the methods for analyzing on-site data for effective training healthcare professional have not been organized. Our study aimed to analyze the effectiveness of this developed system for an individual training. This system was implemented on-site in an actual hospital, through three steps of training: 1) selecting target staff, 2) training target staff and 3) assessing training results. Participants comprised 10 staff members (four members in the treatment and control groups, respectively). Data was analyzed through statistical methods using two indicators: the rate of retaking the blood sample and time taken for blood collection. The results of an effectiveness analysis by analyzing track records, showed a trend of improvement in the retaking rate and the time, but did not show statistical significance through ANOVA and t test for the change in each indicator. In the analysis of the decrease of the most frequent items in retaking situation, there was statistical significance through ANOVA. It showed the plural methods and indicators for verify the effectiveness of individual training.
Condition-based maintenance is preventive maintenance in which an appropriate maintenance action is taken on the basis of the system’s underlying condition before failure. The underlying condition is continuously or periodically observed by a sensor. In most research on condition-based maintenance, the accuracy of the sensor was assumed to not change over time. However, in actuality, not only the system but also the sensor deteriorates over time. In this research, we considered a multi-state system that deteriorates in accordance with a discrete-time Markov chain. It is assumed that the underlying deterioration state is observed periodically by a sensor with accuracy that degrades over time. After each observation, a decision maker chooses one of three actions: continue operating system, replace only system, and replace both system and sensor. We formulated the optimization problem of condition-based maintenance for both the system and sensor using a partially observable Markov decision process. The objective is to minimize the total expected discounted costs (both operation and maintenance of system and sensor) over an infinite horizon. We analytically show that, under certain sufficient conditions, the total expected discounted costs are monotonically non-decreasing with respect to the state probability vector of the system and the age of the sensor. Moreover, the optimal policy has a threshold-type structure. We also provide a numerical example illustrating the properties obtained. These properties are useful for many practical purposes, such as determining thresholds for changing the maintenance action.