Machine Learning Based Intelligent Test Selection for Faster CI/CD Pipelines

CI pipelines become slow as regression suites grow. In many teams, every commit triggers full test execution even when only a few components changed.

In this project, I built a practical prototype that predicts impacted Playwright tests using machine learning.

The Problem

When all tests run on every commit:

feedback is delayed
compute cost increases
developer productivity drops
For large systems, this creates a release bottleneck.

The Idea

Use historical data from CI:

changed files in commit
tests that were impacted (failed, flaky, or behaviorally affected)
Train a model that maps file-change patterns to impacted test files.

Then in CI:

detect changed files
predict relevant tests
run only selected tests first
keep full-suite fallback/nightly run for safety
Example

Commit touches:

src/services/inventory.js
Model predicts:

tests/playwright/tests/inventory.spec.js
tests/playwright/tests/order.spec.js
This gives much faster feedback compared to running all tests.

Tech Stack

Playwright for test execution
Python + scikit-learn for model training/inference
GitHub Actions for CI integration
Implementation Summary

The repository includes:

synthetic commit-impact dataset generator
multi-label classifier (OneVsRest + LogisticRegression)
prediction utility with threshold and safe fallback
CI script that exports SELECTED_TESTS
Playwright runner that executes just selected spec files
Why This Matters

Intelligent test selection is a practical way to improve CI throughput. With good historical data and conservative fallback strategy, teams can achieve significant speedups while preserving confidence.

In many repositories this can reduce per-commit test time by 70-80%.

Repository

GitHub - intelligent-test-selection-ml

If you want, I can share next steps for production hardening (coverage guards, risk bands, retraining cadence, and drift monitoring).