GLOBEM EDA From Raw Sensors to Model Ready Decisions

This report presents a comprehensive exploratory data analysis (EDA) of the GLOBEM dataset to assess its suitability for personalized behavioural trajectory modelling. Using passive smartphone sensing, wearable device data, and mental health surveys collected across four annual cohorts (2018–2021), the analysis evaluates data quality, temporal coverage, missingness patterns, feature distributions, behavioural variability, cross-modal relationships, and cohort drift. The findings provide practical recommendations for designing transformer-based AI models that learn individual behavioural baselines and detect meaningful deviations over time, supporting future research in digital phenotyping and mental health monitoring.本报告针对GLOBEM数据集展开全面的探索性数据分析（Exploratory Data Analysis, EDA），以评估其在个性化行为轨迹建模领域的适配性。本次分析依托2018至2021年间四个年度队列采集的被动式智能手机感知数据、可穿戴设备数据以及心理健康调查问卷数据，对数据集的数据质量、时间覆盖范围、缺失模式、特征分布、行为变异性、跨模态关联以及队列漂移情况展开评估。本研究结果为设计可学习个体行为基线、并能及时识别有效偏差的基于Transformer的人工智能模型提供了实用建议，可为数字表型与心理健康监测领域的后续研究提供支撑。