New ML models improve geotechnical strength prediction with advanced imputation

Researchers have developed new probabilistic indirect models for predicting undrained shear strength in geotechnical engineering, addressing challenges of significant data missing and variability. The study utilized the CLAY/10/7490 global database and tested three imputation methods: multivariate normal (MN), multiple imputation by chained equations (MICE), and miss forest (MF). A Probabilistic Extreme Gradient Boosting (PXGB) model was employed to evaluate the imputation effectiveness. Additionally, a multi-head attention (MHA) mechanism was integrated into an artificial neural network (ANN) to create MHA-based probabilistic neural networks (MHA-PNN), which demonstrated superior performance in prediction accuracy and uncertainty quantification compared to conventional models. AI

IMPACT Enhances prediction accuracy and uncertainty quantification in geotechnical engineering, particularly for datasets with missing information.

RANK_REASON The cluster contains an academic paper detailing new machine learning techniques and models for a specific scientific application. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• artificial neural network
• Atterberg limits
• CLAY/10/7490
• MHA-based probabilistic neural networks
• MHA-PNN
• miss forest
• multi-head attention
• multivariate normal distribution
• piezocone cone penetration
• Probabilistic Extreme Gradient Boosting
• PXGB

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →