Research and application of intelligent prediction of slope stability using an MOIRMO-RF model

A multi-objective Improved Radial Movement Optimized Random Forest (MOIRMO-RF) is introduced to address the issues of slow convergence and overfitting in slope stability prediction. Hyperparameter tuning is formulated as a bi-objective search that jointly maximizes accuracy (Acc) and recall (Rec), with a Pareto archive maintained during global exploration. A comprehensive evaluation adopts the Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) using Acc, precision (Prec), Rec, F1-score (F1), the area under the ROC curve (AUC) and the Average Precision (AP). The dataset comprises 792 literature-compiled cases with six inputs: slope height ([Formula: see text]), slope angle ([Formula: see text]), unit weight ([Formula: see text]), cohesion ([Formula: see text]), internal friction angle ([Formula: see text]), and pore-pressure ratio ([Formula: see text]). Comparative experiments cover MOIRMO-RF, RF, MOPSO-RF, XGB, MOPSO-XGB, MOIRMO-XGB, MLP, MOIRMO-MLP, and MOPSO-MLP. On the test set, MOIRMO-RF achieves Acc = 0.917, Rec = 0.976, Prec = 0.923, F1 = 0.949, AUC = 0.966 and AP = 0.991, delivering concurrent gains in Rec and AUC without compromising precision, thereby improving discrimination. Under multi-objective optimization, convergence is accelerated, generalization is enhanced, and overfitting is suppressed. External validation of mining slope cases from the Yellow River Basin demonstrates robustness and practical usability, supporting applications in risk assessment and early warning decision-making.