Jeon-Insu/V3-gemma: /V3-gemma: V3-Gemma v1.0.1
V3-gemma — Multimodal Feature Extraction and ML Analysis Pipeline Summary This release provides a reproducible end-to-end pipeline for extracting multimodal clinical features (I1–I20) with Gemma 3n and analyzing depression-related signals with machine learning. The repository is organized into four packages that follow the analysis workflow in order. Packages included Multimodal_feature_extraction Extracts clinical features I1–I20 from video (frames + audio) and text using Gemma 3n, with threshold post-processing and paper-ready wide CSV outputs. In-house_machine_learning_toolbox Interactive Streamlit-based ML analysis engine (stratified K-fo
Resource information
- Format
- zip
- Original repository
- 21543711
- DOI
- 10.5281/zenodo.21543711
- Credit basis
- Creators in Zenodo record metadata
- Access
- Open files available from the original Zenodo deposit
- Last checked
- Nov 5, 2026
- License
- cc-by-4.0
- Last record update
- Nov 5, 2026
About this resource
V3-gemma — Multimodal Feature Extraction and ML Analysis Pipeline Summary This release provides a reproducible end-to-end pipeline for extracting multimodal clinical features (I1–I20) with Gemma 3n and analyzing depression-related signals with machine learning. The repository is organized into four packages that follow the analysis workflow in order. Packages included Multimodal_feature_extraction Extracts clinical features I1–I20 from video (frames + audio) and text using Gemma 3n, with threshold post-processing and paper-ready wide CSV outputs. In-house_machine_learning_toolbox Interactive Streamlit-based ML analysis engine (stratified K-fold CV, ROC/AUC, model export, SHAP / permutation importance). Adapted from kskim1105/machine_learning_toolbox by K. Kim. Feature-based analysis_ML Feature-level training and evaluation on extracted I1–I20 features (I4-included setting), including held-out testing and BMI / age subgroup analyses. Criterion_kfold_analysis_ML Criterion-level (Y1–Y9) modeling with stratified k-fold cross-validation, seed-fixed paper reproduction, DSM-5-style voting, and subgroup prediction exports. Intended use Reproduce multimodal feature extraction and downstream ML experiments described in the associated study Train / evaluate feature-level and criterion-level models Generate performance tables, ROC/calibration plots, and subgroup predictions Notes Documentation and code comments are provided in English. Runtime data-label strings required for original Korean metadata matching are preserved where necessary. Please cite the upstream ML toolbox when using the in-house analysis engine. Citation If you use this software, please cite this Zenodo release and the upstream toolbox: This repository / Zenodo DOI (to be assigned) K. Kim, machine_learning_toolbox: https://github.com/kskim1105/machine_learning_toolbox Full Changelog : https://github.com/Jeon-Insu/V3-gemma/compare/v1.0.0...v1.0.1