The estimation of conditional average treatment effects (CATEs) is an important topic in many scientific fields. CATEs can be estimated with high accuracy if data distributed across multiple parties are centralized. However, it is difficult to aggregate such data owing to confidentiality or privacy concerns. To address this issue, we propose data collaboration double machine learning, a method for estimating CATE models using privacy-preserving fusion data constructed from distributed sources, and evaluate its performance through simulations. We make three main contributions. First, our method enables estimation and testing of semi-parametric CATE models without iterative communication on distributed data, providing robustness to model mis-specification compared to parametric approaches. Second, it enables collaborative estimation across different time points and parties by accumulating a knowledge base. Third, our method performs as well as or better than existing methods in simulations using synthetic, semi-synthetic, and real-world datasets.

Small and medium enterprises (SMEs) play a crucial role in supporting the global economy by contributing significantly to employment and value creation. Therefore, accurately predicting early signs of financial distress in SMEs and taking timely management improvement actions is of paramount importance. However, there is a lack of research on developing multi-class financial distress prediction (MFDP) models specifically for SMEs. Moreover, traditional MFDP models often lack interpretability. To address this, this study proposes definitions for multi-class financial distress in SMEs and constructs a MFDP model using machine learning. Additionally, it introduces an explainable framework to interpret the constructed model. Specifically, the study classifies SMEs' financial conditions into three categories: healthy, mild financial distress, and severe financial distress. It conducts a comparative analysis of six machine learning models. Furthermore, it proposes new interpretation methods for the MFDP model using SHapley Additive exPlanations (SHAP) and counterfactual explanations (CE), both of which are explainable artificial intelligence techniques. The empirical results reveal that the MFDP model, which adjusts data balance using random oversampling and integrates LightGBM with a one-versus-rest decomposition method, demonstrates the highest performance. Moreover, the proposed explainable framework demonstrates that it can provide practical and concrete improvement strategies to the model users, such as managers and financial institutions.

The analysis of high-dimensional tensor data presents significant challenges for traditional methods such as CANDECOMP/PARAFAC (CP) decomposition and Higher-Order Singular Value Decomposition (HOSVD) due to the size, complexity, and noise inherent in the data. Recent studies have emphasized the critical need for noise-robust approaches capable of reducing dimensionality while preserving data integrity, particularly for matrix-form data, underscoring the necessity of theoretical advancements in noise reduction. In this study, we propose a novel method, Noise-Reduced Higher-Order Singular Value Decomposition (NR-HOSVD), which integrates noise reduction techniques into the HOSVD framework to enhance dimensionality reduction for matrix data. In comparative simulations with CP decomposition and HOSVD, NR-HOSVD consistently demonstrated lower relative errors as the rank increased, particularly for high-dimensional tensors, surpassing the performance of conventional methods beyond a certain rank threshold. Furthermore, NR-HOSVD exhibited superior stability and faster convergence on large-scale datasets compared to existing approaches. These findings suggest that NR-HOSVD provides a more accurate and efficient approach to high-dimensional tensor data analysis, making it particularly well-suited for applications in big data and machine learning. In the future, NR-HOSVD is expected to serve as a pivotal method for enhancing the accuracy and efficiency of high-dimensional tensor data analysis, with broad applicability anticipated in fields such as large-scale data processing and precision medicine.

Double-entry bookkeeping data (DBD) systematically record the daily transactions of an organization and are the foundation of accounting information. Typically, they are more detailed and larger in scale than financial statements and feature cross-references between accounts. These characteristics make them beneficial for management. However, studies on quantitative evaluations of their usefulness in organizational management are scarce. This study investigates whether features derived from DBD by leveraging their characteristics can enhance future cash-balance predictions, with the primary focus on enhancing the cash management of organizations. To achieve this, DBD obtained from seven medical clinics were represented as graphs using their cross-references, and the graphs were then mapped into vector features based on graph similarity. The experimental results revealed that a long short-term memory model incorporating features unique to DBD obtained higher predictive accuracy than a simple long short-term memory model that only relies on past cash-balance time-series data. Additionally, an auto regressive integrated moving average model was more suitable for some clinics than the long short-term memory model. This can be attributed to the specific business characteristics and structural changes over time within each clinic. The main contribution of this study is the proposal and demonstration of a method that utilizes DBD, which are stored within organizations but often underutilized, to enhance cash management, thereby illustrating their significance. Based on our findings, business intelligence that provides predictive information using DBD could be developed by management professionals and companies offering accounting services.

In recent years, the development of technologies for causal inference with privacy preservation of distributed data has gained considerable attention. Many existing methods for distributed data focus on resolving the lack of subjects (samples) and can only reduce random errors in estimating treatment effects. In this study, we propose a data collaboration quasi-experiment (DC-QE) that resolves the lack of both subjects and covariates, reducing random errors and biases in the estimation. Our method involves constructing dimensionality-reduced intermediate representations from private data from local parties, sharing intermediate representations instead of private data for privacy preservation, estimating propensity scores from the shared intermediate representations, and finally, estimating the treatment effects from propensity scores. Through numerical experiments on both artificial and real-world data, we confirm that our method leads to better estimation results than individual analyses. While dimensionality reduction loses some information in the private data and causes performance degradation, we observe that sharing intermediate representations with many parties to resolve the lack of subjects and covariates sufficiently improves performance to overcome the degradation caused by dimensionality reduction. Although external validity is not necessarily guaranteed, our results suggest that DC-QE is a promising method. With the widespread use of our method, intermediate representations can be published as open data to help researchers find causalities and accumulate a knowledge base.

Misdelivery in logistic services leads to increased costs and degradation of packages. For deliveries using trucks, erroneous deliveries are prevented by checking identification numbers on packages as the packages pass through delivery points. In recent years, it has become possible to monitor real-time location by attaching GPS receivers to packages, but few concrete efforts have been made to detect anomalies using truck transport data. This study introduces a basic framework for the detection of misdeliveries and a summary of the problems in actual truck misdelivery using special medical supply delivery data provided by a major Japanese logistics company. It is shown that the system can detect erroneous deliveries with high accuracy, even for actual delivery data with coarse resolution owing to the cost of installing the equipment. This study has the potential not only to improve logistics and reduce costs, but also to solve various social problems such as driver shortages.

In recent years, an increasing number of studies have attempted to extract information from journals that cannot be conveyed in financial statements. Journals are valuable assets that enhance corporate value and companies’ competitiveness because they contain abundant information about past transactions. However, only a few studies have quantitatively clarified the value of double-entry bookkeeping and journals. Therefore, finding useful business knowledge in journals is currently difficult for managers and administrators. This study focuses on cross-reference information, which is unique to double-entry bookkeeping and not included in financial statements and trial balances. We then created new features to enhance the explanatory power of future business performance from journals. To achieve this objective, we constructed a graph based on journals, in which cross-reference is represented as edges. We create features unique to journal entry using two methods: network metrics and feature generation by node embedding. Finally, we tested for Granger causality using a vector autoregressive model constructed from created feature and performance. Our experiment showed that we successfully identified Granger causality within the unique features of double-entry journals for all five performance indicators of seven companies. Practitioners can use the identified features to build an alert system for sudden performance declines and facilitate understanding of their business. Future research will be necessary to verify the significance of double-entry bookkeeping and journal entries.

In recent years, rapid digitization and technological advances have enabled governments and companies to utilize various types of personal information, including environment, medical, and lifestyle. Privacy protection is an important issue in the utilization of personal information, and anonymization methods such as k-anonymization and 1-diversity have been proposed in the past. However, to utilize data effectively, it is also important not to lose the usefulness of the information. Privacy protection and preserving the usefulness of information are in a trade-off relationship, and balancing these two aspects has been a difficult problem. Recently, microaggregation using clustering algorithms has been attracting attention as an anonymization method that ensures the usefulness of the data while protecting privacy. In this study, we proposed a new microaggregation method based on iterative processing using probabilistic latent semantic analysis and evaluated the proposed method based on the use case. The use case is a classification of social isolation and an analysis of the relationship with physical frailty risk using the classified groups. Compared to microaggregation using quasi-identifiers and microaggregation based on a conventional clustering algorithm, the proposed method was able to find social isolation characteristics obtained in the raw data with higher accuracy. In addition, in a more detailed analysis focusing on a population with certain characteristics, the proposed method was able to replicate the analysis results obtained using the raw data and to make appropriate assessments of the health risks.

In this study, we investigated whether visual and tactile symmetrical stimuli in the arm affect the mirrored avatar’s sense of ownership of the hand. In the experiment, we tested the user, avatar, and mirrored avatar’s sense of ownership of their hand by catching a ball multiple times in a virtual space. The results suggest that non-ambidextrous individuals were significantly more likely to recognize the mirror-image avatar’s hand as their own in scenarios that included symmetrical visual and tactile stimuli than in scenarios without such stimuli. This experiment may contribute to a better understanding of the relationship between left-right sensation and body ownership sensation. It may contribute to the pursuit of immersive experiences in virtual environments.

We investigate the impact of voice imitation on rapport building in customer service in a virtual environment(VE). We simulated a VE store and conducted a within-subject experiment in three customer service scenarios with 16 participants. The imitation condition used a voice generated by a machine learning model to reproduce the participant’s linguistic content, with the voice identity set midway between the participant and the salesperson. In a group of men, voice imitation significantly improved their impression of salespeople. The findings of this study can be used to design interpersonal services in VE with the help of salesperson avatar voice design.

The opportunities to receive and provide services on virtual environment (VE) continue to increase. In this study, we focused on the voice in VE, and investigated the effects of whisper voice, which is used in intimate relationships, on interpersonal distance between salesperson and customer and the effective range of ventriloquism effect. The experimental results showed two results when the salesperson used whisper voice. The interpersonal distance between them was significantly smaller than that of the normal voice. The effective range of the ventriloquism effect was significantly larger on the front side and significantly smaller on the backside compared to normal voice.

Recently, data collaboration (DC) analysis has been developed for privacy-preserving integrated analysis across multiple institutions. DC analysis centralizes individually constructed dimensionality-reduced intermediate representations and realizes integrated analysis via collaboration representations without sharing the original data. To construct the collaboration representations, each institution generates and shares a shareable anchor dataset and centralizes its intermediate representation. Although, random anchor dataset functions well for DC analysis in general, using an anchor dataset whose distribution is close to that of the raw dataset is expected to improve the recognition performance, particularly for the interpretable DC analysis. Based on an extension of the synthetic minority over-sampling technique (SMOTE), this study proposes an anchor data construction technique to improve the recognition performance without increasing the risk of data leakage. Numerical results demonstrate the efficiency of the proposed SMOTE-based method over the existing anchor data constructions for artificial and real-world datasets. Specifically, the proposed method achieves 6, 4, and 36 percentage point performance improvements regarding NMI, ACC and essential feature selection, respectively, over existing methods for an income dataset. The proposed method provides another use of SMOTE not for imbalanced data classifications but for a key technology of privacy-preserving integrated analysis.

In this study, we constructed a model for predicting the future frailty status of individuals based on information regarding their current lifestyle habits. In recent years, as the global population ages, the number of people certified for long-term care has increased. In an effort to prevent the need for long-term care, prediction of the onset of frailty, which is positioned as a preliminary stage of the caregiving state, is gaining attention. However, the onset of frailty takes a long time, and the incidence of frailty differs according to one’s health literacy. Therefore, a prediction is extremely difficult to achieve. In this study, we constructed a prediction model using a probabilistic latent semantic analysis and a Bayesian network. First, the probabilistic latent semantic analysis was conducted to classify individuals according to their health literacy. Second, we constructed a future frailty prediction model using a Bayesian network for each health literacy segment. As a result, we clarified the following: It is more appropriate to predict a future frailty by predicting the future lifestyle using information on the current lifestyle. In addition, more accurate prediction models can be constructed when individuals are divided based on their health literacy.

This paper proposes an interpretable non-model sharing collaborative data analysis method as a federated learning system, which is an emerging technology for analyzing distributed data. Analyzing distributed data is essential in many applications, such as medicine, finance, and manufacturing, due to privacy and confidentiality concerns. In addition, interpretability of the obtained model plays an important role in the practical applications of federated learning systems. By centralizing intermediate representations , which are individually constructed by each party, the proposed method obtains an interpretable model, achieving collaborative analysis without revealing the individual data and learning models distributed between local parties. Numerical experiments indicate that the proposed method achieves better recognition performance than individual analysis and comparable performance to centralized analysis for both artificial and real-world problems.

In this study, we propose a sensor network to estimate service reassurance in a store by estimating customer stress level based on a pulse wave. We developed a basket that can calculate an RR interval (RRI) with sufficient accuracy to estimate customer reassurance from the measured pulse wave under in-store conditions. Two processes were applied to the measurement values based on the assumption that certain circumstances may interfere with measuring pulse waves. Furthermore, the authors conducted pulse wave measurement experiments under simulated store conditions. The results showed RRIs' values with an error rate of 4.9% compared to the true value.

Although a significant volume of research about sensing networks are conducted to measure environmental distribution changes mainly in outdoor used GPS positioning techniques, there are few indoor positioning systems collecting information other than position. In this research, we developed the sensor network platform for customer behavior analysis in retail stores. These performances are confirmed through some experiments, and the increase or decrease of the goods, which are put into the baskets, is detected through the sensor network of IR signals.