Glucose Screening
Description
This dataset provides 15,720 measurement sessions for non-invasive diabetes screening research, combining three data modalities: dual-wavelength near-infrared (NIR) optical sensor readings (660nm and 940nm, plus a derived 660/940nm ratio), wearable physiological signals (heart rate, blood volume pulse, electrodermal/skin conductance, wrist temperature, and movement intensity), and demographic variables (age, BMI). Each session is labeled final_label__is_diabetic (1 = diabetic, 0 = non-diabetic), derived from a paired blood glucose measurement using the standard 126 mg/dL clinical threshold. The dataset aggregates 57 columns from five underlying sources — PhysioCGM, Nature Scientific Reports NIR Glucose, Kaggle Raman Diabetes, Raman Sugars, and NTNU NIR Glucose — combining 56.1% real-world sensor recordings with 43.9% statistically modeled synthetic data generated to be consistent with published correlation values. Class distribution is 66.7% diabetic and 33.3% non-diabetic. The data is pre-partitioned into a stratified 70/15/15 train/validation/test split (identified by a split column), preserving class ratio across partitions. The dataset is intended to support development of non-invasive diabetes screening models — i.e., predicting diabetes risk from wearable- and optical-sensor signals alone, without a blood draw. Exploratory analysis shows the NIR optical features carry the strongest signal (Pearson r up to −0.72 with the diabetic label), followed by heart-rate-derived wearable features (moderate correlation), while demographic features (age, BMI) show negligible correlation in this specific dataset. A full data dictionary, per-feature correlation/statistical significance tables, and the code used to generate all summary tables and figures are included.
Files
Steps to reproduce
1) Download the dataset CSV file(s) included with this deposit. 2) Open the accompanying notebook "Multimodal_Diabetes_Dataset_EDA_DataInBrief.ipynb" in Jupyter Notebook or Google Colab. 3) Run Step 0 to import required libraries (pandas, numpy, matplotlib, seaborn, scipy). All libraries are standard and pre-installed in Google Colab; no additional installation is required. 4) Run Step 1 to load the dataset CSV file (or a ZIP file containing it). 5) Run the Column Auto-Detection step. This automatically identifies the relevant column groups (target label, source dataset, glucose value, and Tier 1/Tier 2/Tier 3 feature groups) by keyword matching and prints the matched columns for verification. 6) Run Steps 2 through 17 sequentially to reproduce the full analysis, including: data preview; dataset shape and column category breakdown; data type summary; missing value and duplicate checks; an auto-generated data dictionary; descriptive statistics (overall and grouped by class); the dataset composition dashboard (class balance, real vs. synthetic split, source dataset breakdown, and train/validation/test split); Pearson correlation analysis and feature tiering; univariate distribution plots; bivariate scatter plots; group mean comparison with independent-samples t-tests and Cohen's d effect size; IQR-based outlier detection; cardinality checks; and internal consistency checks (including verification of the diabetic label against the 126 mg/dL glucose threshold). 7) Run Step 18 to export all generated figures (PNG format) and tables (CSV format) to the working directory, including the data dictionary, summary statistics, and correlation results. All numerical results reported in the accompanying report, "Diabetes Prediction from Non-Invasive Body Signals," including headline counts, per-feature correlations, and group means, are computed directly within this notebook and can be traced to their corresponding code cell.