FlyRank — Search Performance Decline Prediction & SEO Prioritization
ML-based decision-support system for identifying webpages with potential search-performance decline and prioritizing SEO/content reviews using historical search-performance signals.
Architecture
Large-scale search-performance data processing (~78.8M records)
Target-leakage detection & feature engineering
Client-grouped 5-fold cross-validation scheme
Model family evaluation (Logistic Regression, Random Forest, XGBoost, LightGBM, CatBoost)
Human-in-the-loop content prioritization action workflow
Results
Random Forest achieved Precision@50 of 0.444 compared with a 0.392 baseline (+5.2 percentage points).
Developed client-grouped cross-validation to prevent data leakage across domain clients.
Challenges
Handling data leakage across client-grouped domains
Imbalanced search decline target distribution
Evaluating ranking precision vs standard classification metrics