user@devops:~$ cat README.md
Model Drift Detection
# Description
Advanced MLOps project simulating a real production scenario with progressive drift. A Random Forest is trained on 10,000 samples with 6 features (age, income, credit score, transaction count, account tenure, usage rate) and evaluated against 10 production batches. Early batches have only data drift (feature mean and variance shifts), batches 5+ add concept drift (credit score's target relationship reverses). 4 detection metrics: PSI (>0.25 = severe drift), KS test (p<0.05), accuracy decay, and prediction distribution shift. 6 visualizations including PSI heatmap, temporal evolution, multi-metric dashboard, and distribution comparisons.
# Key features
$ Data Drift: shifts in mean, variance, and distribution shape
$ Concept Drift: gradual reversal of a feature's target weight
$ Population Stability Index (PSI): thresholds 0.1 (mild) and 0.25 (severe)
$ Kolmogorov-Smirnov test: per-feature detection with p<0.05
$ Accuracy decay tracking: 95.5% baseline, drop to 88.7% at batch 6
$ 6 visualizations: PSI evolution, accuracy decay, heatmap, distributions, dashboard
# Gallery
# Technologies used
