A Method for Predicting Lithium Associated with Aluminum-bearing Rock Series in Wuzhengdao Area of Northern Guizhou Based on Machine Learning

Lithium （Li） in aluminum-bearing rock series is mainly characterized by "scarcity"， "association" and "fineness"， and these features lead to complex geochemical behaviors， making it challenging to predict enrichment areas. Machine learning （ML） technologies provide an effective way to solve the above problems. This study employs Random Forest （RF）， eXtreme Gradient Boosting （XGBoost）， Light Gradient Boosting Machine （LightGBM）， and Categorical Boosting （CatBoost） to build regression models with major elements as input features. Evaluation results identify CatBoost as the best-performing model. Using the SHapley Additive exPlanations （SHAP） method and feature subset selection， Al₂O₃， SiO₂， MgO and K₂O were confirmed as the primary predictive factors. The CatBoost model demonstrated robust generalization ability and effectively predicted the variation pattern of Li content at the profile scale， confirming its practical application value. After applying the Synthetic Minority Over-sampling Technique for Regression （SMOTER） algorithm， the CatBoost model showed markedly enhanced predictive accuracy for high-content samples. Notably， the model maintained reliable performance against ±15% variations in major element data， underscoring its suitability for engineering applications in mineral exploration. This study thus provides a rapid and cost-effective method for Li exploration in aluminum-bearing rock series.