580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can variable selection and feature importance methods enhance Random Forest Regression accuracy and interpretability in high-dimensional or complex data?
This research area focuses on identifying optimal variable subsets and measuring feature importance in Random Forest Regression (RFR) models to improve prediction accuracy and interpretability, especially when dealing with high-dimensional data or when predictor variables are correlated, categorical, or of mixed types. Understanding and managing variable selection reduces noise, limits bias, and enhances model generalization.
Key finding: This study presents a stepwise Random Forest (SRF) variable selection method that outperforms standard variable selection techniques such as Boruta, VSURF, and linear stepwise regression for predicting forest growing stem... Read more
Key finding: Introduces the Intervention in Prediction Measure (IPM), a novel variable importance measure for Random Forests independent of prediction performance metrics and adaptable to multivariate responses. IPM, based on tree... Read more
Key finding: Proposes xRF, an improved Random Forest algorithm incorporating unbiased feature sampling by separating informative from uninformative features using p-value and chi-square tests before splitting. This approach addresses bias... Read more
Key finding: Demonstrates that pre-estimation dimension reduction (targeting) via supervised variable pre-selection (e.g., LASSO) enhances Random Forest Regression performance by increasing the probability of splits on strong predictors,... Read more
Key finding: Develops a methodology to select significant variables within Random Forest classification models applied to chemical data (NMR spectra) to interpret the influence of variables on maximum pour point (MPP) of crude oil. Using... Read more
2. How can Random Forest Regression be adapted or combined with ensemble and optimization techniques to improve predictive speed and accuracy in real-world applications?
This theme investigates the development of ensemble variants, pruning methods, and hybrid frameworks of Random Forest Regression (RFR) to optimize computational efficiency and maintain or improve predictive accuracy. It addresses practical constraints in healthcare, environmental modeling, manufacturing, and sensor-based systems where faster inference or better generalized performance is critical.
Key finding: Introduces CLUB-DRF, a pruned Random Forest ensemble that clusters similar trees to inject diversity and select representatives, resulting in a substantially smaller model (over 92% pruning) with equivalent or improved... Read more
Key finding: Proposes a novel soft sensor combining Random Forest Regression (RFR) and Partial Least Squares (PLS) for dynamic process modeling, showing improved one-step-ahead prediction accuracy and stability over traditional models... Read more
Key finding: Develops RFR models to surrogate computational fluid dynamics (CFD) simulations of turbulent flow characteristics in curved pipes, producing significant computational cost reductions while maintaining high prediction accuracy... Read more
Key finding: Combines Random Forest with Random Search optimization (RS-RF) to predict soil erosion status, achieving improved classification metrics (accuracy, MCC, F1-score) by optimizing RF hyperparameters via metaheuristics. The study... Read more
Key finding: Presents hybrid models coupling Random Forest Regression with metaheuristic optimizers FDA, GJO, and GTO for predicting maximum dry density of soil. The RFGJ (RF with GJO optimizer) model achieved the highest R² (0.9966) and... Read more
3. What are the practical applications of Random Forest Regression in diverse domains such as healthcare, environmental monitoring, remote sensing, and manufacturing for accurate and interpretable predictive modeling?
This theme highlights the application of Random Forest Regression (RFR) in real-world, domain-specific problems where accurate prediction and model interpretability are necessary. It surveys how RFR supports decision-making in healthcare systems, hydrology, mobile network performance, manufacturing machining processes, and ecological modeling, showing its adaptability across multidisciplinary datasets.
Key finding: Demonstrates that Random Forest Regression, trained on remotely sensed data, can predict streamflow in a snowmelt-dominated mountainous watershed with higher accuracy and less calibration effort compared to the Soil and Water... Read more
Key finding: Develops an IoT-enabled healthcare system using sensors connected via Raspberry Pi and employing Random Forest Regression models to accurately predict physiological measures such as heart rate and blood pressure. The system... Read more
Key finding: Applies multiple machine learning models including Random Forest Regression to predict mobile network throughput performance (downlink bit rate) based on cellular network parameters. The Random Forest model obtained highest... Read more
Key finding: Assesses Random Forest Regression performance on count data sets with overdispersion, comparing it with classical count-based generalized linear models. Results indicate RF achieves comparable or better predictive accuracy,... Read more
Key finding: Evaluates machine learning algorithms including Random Forest Regression to model the relationship between machining process parameters and surface roughness. RFR's ensemble learning capability effectively captures... Read more