Predicting Microbiologically Influenced Corrosion Risk from Quorum Sensing Biofilm Community Features: A Random Forest-SHAP Approach

Abstract

Bipul Bhattarai, Sulav Dahal and Naina Maharjan

TMicrobiologically influenced corrosion (MIC) causes an estimated $30–50 billion in annual infrastructure losses across oil and gas pipelines, marine systems, and water distribution networks. To our knowledge, no prior machine learning model has explicitly incorporated quorum sensing (QS) signals — the bacterial cell-to-cell communication system governing biofilm formation and maturation — as predictive features for MIC risk assessment. We propose a Random Forest-SHAP framework integrating QS biofilm community features (QS activity score, AHL signal proxy, QS-biofilm synergy index) alongside environmental parameters (dissolved oxygen, pH, sulfate, H2S) and microbial community composition (sulfate-reducing bacteria abundance, biofilm aggression score) to classify MIC risk as high or low (threshold: 50 μm/year). A curated dataset of 78 experimental records was assembled through semi-automated literature extraction followed by manual verification from more than 15 published MIC studies. Random Forest achieved crossvalidated F1=0.762 and AUC-ROC=0.846, outperforming XGBoost, SVM, and MLP baselines. SHAP analysis identified dissolved oxygen, temperature, and chemical stress score as top predictors, with QS-derived composite features contributing to MIC risk stratification.

PDF