Dr-P/amy-commitment-analysis

Organizational Commitment Analysis — Research Portal

A full research portal for explainable ML analysis of organizational commitment (Stay vs. Leave). Designed for publication-ready analysis with complete audit trail.

Features

Data & Preprocessing

• —Upload Excel (.xlsx/.xls) or CSV files
• —Dynamic column selection — choose features, target, and categorical vs numeric columns
• —Preprocessing pipeline: imputation, scaling, feature selection, class balancing

Machine Learning Models

• —Logistic Regression, Random Forest, XGBoost, SVM (RBF), PyTorch MLP
• —Hyperparameter controls with per-model sliders
• —Auto-tune via Optuna (Bayesian) or GridSearchCV fallback
• —5-fold stratified cross-validation with F1 scoring

Explainable AI

• —Permutation Feature Importance (PFI) — model-agnostic, all models
• —SHAP values — TreeExplainer for tree models, KernelExplainer for others
• —LRP (Layerwise-Relevance Propagation) — epsilon and alpha-beta rules on the MLP
• —Logistic Regression coefficients — log-odds with direction

Portal Tools

• —Google Drive resource library integration
• —Claude-powered Code Assistant — modify source files via natural language
• —Full equations reference with LaTeX-rendered math and citations
• —Append-only Decision Log with complete audit trail, exportable as .txt

Independent Variables

• —Gender (M / F / NB)
• —Education (HS → PhD, ordinal)
• —Ethnicity (5 categories)
• —Marital Status (Single / Married / Divorced / Widowed)
• —Reason for Joining (Career / Service / Family Tradition / Education Benefits / Camaraderie)
• —Ideology (1–7 Likert)

Dependent variable: Commitment — Stay (1) or Leave (0)

Tabs

1. 1.Home / Resources — welcome, instructions, Google Drive folder
2. 2.Data — load synthetic (N=300) or upload your own Excel/CSV; select columns
3. 3.Preprocessing — imputation, scaling, feature selection, class balancing
4. 4.Model Config & Train — hyperparameter sliders, auto-tune, train all models
5. 5.Performance — accuracy, F1, AUC, confusion matrices
6. 6.Explainability — PFI, SHAP, LRP (mean + per-sample), LogReg coefficients
7. 7.Equations & Rationale — full math with citations (Breiman 2001, Bach 2015, Montavon 2019)
8. 8.Code Assistant — read and modify source files via Claude API
9. 9.Decision Log — timestamped audit trail, exportable

File Upload Format

Any Excel/CSV with at least one numeric or categorical feature column and a binary target. Built-in synthetic dataset requires no upload.

Environment

For the Code Assistant tab, set ANTHROPIC_API_KEY in a .env file or as a HF Space Secret.