深澤 佑介

Abstract In this paper, we propose a hybrid approach that combines Small Language Model (SLM)-based interpretation with machine learning (ML)-based prediction to analyze stress levels and related factors from step-count data. While several datasets exist for predicting mental health conditions from sensor data, most do not explicitly address the underlying factors associated with stress. To explore this issue, we collect step-count data from 30 nurses, together with stress assessments (QIDS: Quick Inventory of Depressive Symptomatology) and stress factor ratings based on six questionnaire items measured on a 4-point Likert scale, collected over 8 days within 4 weeks. We evaluate the proposed approach through two tasks. The first task examines how intermediate textual interpretations relate to stress presence estimation. Under our baseline experimental settings, BERT (Bidirectional Encoder Representations from Transformers) with intermediate stress interpretations achieved the highest accuracy (0.74), compared with BERT using raw step-count representations (step count: 0.63, distance: 0.59) and a prompt-based approach. The second task evaluates the association between intermediate interpretations and stress factor ranking. In this setting, BERT with intermediate stress interpretations achieved a ranking accuracy of 0.60, compared to 0.56 when using step-count sequences without interpretation. Higher correlations were observed for work-related stress factors such as “workplace relationships,” “busy work,” “heavy work responsibilities,” and “lack of time off.” Overall, these results suggest that intermediate textual representations derived from step-count data can be useful for stress analysis under baseline conditions, while avoiding causal claims about stress determinants.

近年，バイクシェアサービス（Bicycle Sharing Services; BSS）に注目が集まっている．BSSの利用が増えることで，専用駐輪場（ポート）の自転車溢れ・不足の問題が顕在化している．この問題の解決方法の1つとして自転車が溢れているポートから自転車が不足しているポートへ自転車を移動する再配置がある．再配置業務をより効率的に行うためには正確な自転車の需要を予測することが重要である．そこで，本研究ではBSSの需要をポート・時間ごとに予測する機械学習モデルを提案する．提案手法は，変分オートエンコーダーに時系列生成モデルを組み合わせたモデルである．我々はこの手法を“Conditional Variational Autoencoders considering Partial Future data”（CVAE-PF）と名付けた．我々はオフライン実験・オンライン実験を通じて提案手法の評価を行った．オフライン実験ではRoot Mean Square Error（RMSE）を用いた評価を行い，提案手法の予測精度が比較手法に比べて高いことを示した．また，オンライン実験では従来の再配置方法に比べ，ポートごとの自転車の溢れ・不足を減少させることができることを確認した． In recent years, Bicycle Sharing Services (bike-shares) have been established worldwide. One important aspect of bike-share management is to periodically rebalance the positions of the available bikes. Because the bike demand varies by and over time, the number of bikes at each bike-port tends to become unbalanced. To efficiently rebalance a bike-share system, it is essential to predicting the number of bikes in each bike-port. In this paper, we propose a method to predicting bike demand and the number of bike pickups and drop offs at each bike-port every hour, up to 24 hours in advance. To predict demand, we used a time series generation model based on the Variational Autoencoders model and the Attention based Sequence to Sequence learning model. We named this method “Conditional Variational Autoencoders considering Partial Future data” (CVAE-PF). In the experiment, our proposed method showed higher prediction accuracy in root mean square error (RMSE) compared to conventional methods. In addition, the constructed demand model of bicycle sharing services was used for rebalance, and it was confirmed that rebalance could be carried out more efficiently than with conventional methods.

飲食店において売上などの需要を予測することは，スタッフのシフトスケジューリングや店舗の事前準備を行ううえで重要なタスクである．特に，イベント開催や天候変動を契機として通常より高い売上が発生する高需要状態を事前に予測することが求められている．本研究では，過去の売上データに加え人口統計データを用いることで，高需要状態での予測精度を改善する手法を提案する．提案手法では，売上予測のモデルと，将来需要が高くなるか低くなるかを判断するモデルの2つのモデルの結果を組み合わせて売上予測を行う．飲食店における実売上データを用いて予測精度を評価した．その結果，従来手法と比較して，高需要時の需要予測精度が1.45%向上することを示した． Predicting the future sales volume is a critical task in operating a restaurant as it helps to determine the number of staff and ingredients to have at any time and decide when to start preparing food. Future sales can typically be predicted according to the demand cycle; however, it is difficult to predict an immediate increase in demand because it is out of the demand cycle. To tackle this issue, this study proposes a method for predicting the next-hour future sales volume based on population statistics data, in addition to historical sales and current and historical weather data. The proposed method combines the results of two models, one predicting the sales volume, while the other determining whether the future demand will become high or low. The proposed method was evaluated using actual restaurant data collected in collaboration with a major Japanese restaurant company. The results demonstrate that the prediction accuracy can be improved by 1.45% compared to the prediction model of sales volume without combination of demand classification when the sales volume is high.

昨今のデータ収集，活用の機運の高まりにより，健康診断のデータ化が進められている．本研究では健康診断のデータを用い，健康の維持や向上を実現するための行動変容を目的とした技術として，生活習慣改善行動における価値推定手法を提案する．提案手法では健康診断における生活習慣に関する問診への回答結果，および医師による受診者の総合所見のデータを用い，将来の総合所見を予測するモデルを構築する．さらに，提案手法では構築した総合所見予測モデルに加え，価値計算フレームワークおよび生活習慣改善内容の選定機構を構築し，ユーザがとりうる生活習慣改善行動の価値を定量化し，価値の高い行動を選定する．実際に会社従業員のべ5,000人超の健康診断データに基づき提案手法による予測モデルを構築した結果，各データにおける翌年の医師による総合所見の予測精度が約67.2%であり，総合所見が変わらないと仮定するベースラインの推定方式と比較して7.1ポイント改善できることを確認した．さらに，提案手法を実際のデータに適用することにより，生活習慣の改善において重要な問診項目を複数確認した． With the increasing demand for data collection and utilization in recent years, the digitization of medical examination data is being promoted. In the medical domain, the key challenge of data analysis is shifting from understanding the general health context that can be derived from accumulated data to real-world insights. In this study, we propose a method to value lifestyle improvement actions to help people adopt healthier behaviors. To value and recommend a change of lifestyle, the proposed method consists of three components: prediction model of overall health status, a framework to value a lifestyle change, and a scheme to determine lifestyles to be improved. the valuable advice generation. An evaluation of more than five thousand medical examination data revealed that the prediction model could predict the overall health status in the next year with 67.2% accuracy and boost accuracy by 7.1 points, even if the model only relied on answers to diagnostic questions and the overall health status in the current year.