Chapter Four · failure evidence
What Random Forests & Tree Ensembles got wrong, from 91 dissertations
Tree ensembles and random forests frequently underperform simpler linear baselines or alternative architectures such as gradient boosting and neural networks across diverse tasks. Researchers also encounter severe failures related to overfitting on high-dimensional data, poor calibration on extreme values, class imbalance degradation, and uninterpretable model complexity. These records come from PhD theses at 29 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Tree ensembles underperformed simpler linear and parametric baselines
Across multiple studies, random forests and decision trees achieved lower accuracy or higher error compared to multiple linear regression, logistic regression, and generalized linear models. These models struggled when linear baselines captured the underlying structure better or when tree ensembles failed to generalize on continuous outcomes and spatial surveys.
Tried and failed
generalized random forest heterogeneous treatment effect estimation applied to continuous outcome treatment effect estimation. Outcome: worse than baseline. Reason: generalized random forest performed poorly estimating heterogeneous treatment effects on continuous variables compared to binary outcomes
THREE ESSAYS ON E-COMMERCE DEVELOPMENT AND INEQUALITY IN CHINA · Cornell
Lost to a baseline
Random Forest models exhibited poorer performance and greater subgroup disparities compared to simpler logistic regression models.
Lost to a baseline
Decision tree predictive R2 (0.436 with DRS, 0.260 without for FPC1) and Random Forest (0.479 / 0.370) were lower than Elastic Net (0.550 / 0.450).
Lost to a baseline
Random Forest (ensemble method) performed substantially worse than the simpler single Classification Tree, achieving only 42.43% agreement with BorealDB versus CT's 84.88%.
Lost to a baseline
Random Forest classifier (macro F1-score 0.87, accuracy 0.88) lost to standard Decision Tree classifier (macro F1-score 0.94, accuracy 0.94) on EMG gesture recognition.
Lost to a baseline
Random Forest classifier (AUC=0.707) was beaten by standard logistic regression (AUC=0.752)
Lost to a baseline
Random forest variable selection underperformed simpler stepwise AIC and LASSO regression in variable recovery and discrimination/calibration metrics
Enhancing Survival Prediction Models: Insights on Biomarker Inclusion and Model Updating · JScholarship
Lost to a baseline
Random forest regression models did not significantly outperform multiple linear regression, leading to favoring simpler MLR models.
Lost to a baseline
Random Forest trained on RAVDESS+MESD had lower external validity on TESS than simple Linear Regression and Lasso.
Creating Links: Building an Educational Platform to Ask Relevant Questions in Education · MIT
Tried and failed
random forest classification applied to imbalanced tabular student outcome prediction. Outcome: worse than baseline. Reason: lower AUC and specificity compared to regularised linear models and gradient boosting across imbalanced panels
Using Machine Learning to Advance High School Dropout Prediction and Prevention · Penn
Tried and failed
random forest classification and survival modeling applied to clinical tabular outcome prediction. Outcome: worse than baseline. Reason: complex machine learning models failed to outperform classical logistic regression and Cox proportional hazards baselines
Tried and failed
Random forests and k-nearest neighbors classification applied to optimization oracle indicator function approximation. Outcome: worse than baseline. Reason: Non-linear classifiers yielded poor generalization accuracy compared to linear baselines like logistic regression.
Consistency in integer programming · Iowa State
Tried and failed
decision tree classifiers applied to sparse bag-of-words text classification. Outcome: worse than baseline. Reason: decision trees produced poor and inconsistent classification performance compared to logistic regression
Andromeda in Education: Studies on Student Collaboration and Insight Generation with Interactive Dimensionality Reduction · Virginia Tech
Tried and failed
random forest regression applied to predicting physical parameters from relaxometry data. Outcome: worse than baseline. Reason: non-linear ensemble failed to outperform a simpler quadratic general linear model baseline
Quantitative Microstructural Imaging for Clinical Use · EPFL
Tried and failed
random forest regression replacing linear mixed-effects models applied to spatial land use regression modeling. Outcome: worse than baseline. Reason: provided no improvement in predictive accuracy over linear mixed-effects regression
Spatiotemporal patterns and socioeconomic inequalities of noise and sound sources in Accra, Ghana · Imperial
Lost to a baseline
Random Forest (0.519 accuracy) and Bagging Classifier (0.618 accuracy) underperformed compared to simpler Decision Tree (0.994) and Logistic Regression (0.984).
Measuring and Analyzing Community Resilience During COVID-19 Using Social Media · Virginia Tech
Lost to a baseline
Random forest (AUC = 0.6345) was outperformed by full logistic regression (AUC = 0.65991) on event status prediction.
ENHANCING NAVAL RETENTION: A STRATEGIC APPROACH TO ALLOCATING SELECTIVE REENLISTMENT BONUSES · Calhoun
Lost to a baseline
Random forest on JDTC data achieved 75.00% test accuracy, beaten slightly by standard logistic regression at 77.08%.
Considered and rejected
Considered and rejected: Rejected standard/complex classifiers (SVMs, Random Forests) for Rim Thickness Curves in glaucoma diagnosis because logistic regression performed equally or better
Leveraging Uncertainties in Medical Prediction Systems · Publikationssystem UB Tuebingen
Tried and failed
gradient boosted trees with TF-IDF features applied to short text classification. Outcome: worse than baseline. Reason: linear model handles high-dimensional sparse TF-IDF text features better than decision trees
Urban Housing in the Digital Age: Three Applications of Natural Language Processing · Cornell
Tried and failed
random forest regression on functional connectivity features applied to predicting behavioral retrieval accuracy. Outcome: worse than baseline. Reason: connectivity features in the baseline condition lacked predictive signal, yielding negative explained variance
Tried and failed
random forest for small area estimation applied to hierarchical zero-inflated domain data. Outcome: worse than baseline. Reason: predictors lacked domain-level random intercept information, causing massive bias when domains varied independently of features
A Forest for the Trees: Using Random Forests for Small Area Estimation on US Forest Inventory Data · Harvard
Tried and failed
random forests for small area estimation applied to zero-inflated spatial survey data. Outcome: worse than baseline. Reason: failed to outperform linear and mixed-effects models across varied sample sizes, correlations, and target variable ranges
A Forest for the Trees: Using Random Forests for Small Area Estimation on US Forest Inventory Data · Harvard
Lost to a baseline
Random forest models including full factor sets failed to outperform simpler multiple linear regression models in explaining variance in evolutionary rate or purifying selection.
Random forests suffered from severe overfitting on sparse or high dimensional data
Models fitted training noise and created discontinuous decision boundaries when trained on high-dimensional features or small sample sizes. Consequently, performance dropped sharply on holdout cohorts, validation environments, and pruned feature sets.
Tried and failed
random forest classification applied to high-dimensional sparse genomic variant features. Outcome: overfit. Reason: severe overfitting during training on high-dimensional genomic features compared to logistic regression baseline
Decoding Germline Genetic Influence on Cancer Somatic Mutation Acquisition · Harvard
Tried and failed
random forest classification on low-dimensional biomarker data applied to disease diagnosis prediction. Outcome: overfit. Reason: models formed discontinuous islands and narrow decision bands fitting training noise
Machine-Learning Aided Diagnosis Of Alzheimer's Disease · UT Austin
Tried and failed
random forest regression applied to oceanic tracer distribution modeling. Outcome: overfit. Reason: prone to overfitting on sparse data and underperformed gradient boosting
Environmental pollution as a marine tracer: measuring and modelling anthropogenic lead in the ocean · Imperial
Tried and failed
Random forest regression on dynamic response features applied to crack depth prediction in concrete. Outcome: overfit. Reason: High-dimensional feature space on concrete dynamic response data led to significant overfitting.
Machine learning (ml) approaches to model interdependencies between dynamic loads and crack propagation · Cranfield
Tried and failed
random forest for chemical property prediction applied to mass spectral chromatographic data. Outcome: did not generalise. Reason: pretrained random forest model suffered significant accuracy drop when applied to new experimental chromatographic dataset
A Statistical Methods-Based Novel Approach for Fully Automated Analysis of Chromatographic Data · Virginia Tech
Tried and failed
random forest package implementation for probability estimation applied to discrete choice consideration set formation. Outcome: overfit. Reason: Default implementation overfitted severely, assigning nearly perfect probability to observed alternatives unlike Ranger
Endogeneity and consideration-set issues in residential location choice models · Imperial
Tried and failed
random forest regression applied to tabular spatial feature regression. Outcome: overfit. Reason: Severe overfitting to training data, high computational training time, and poor spatial variance compared to gradient boosting.
Tried and failed
random forest classification without feature selection applied to high-dimensional molecular descriptor dataset. Outcome: overfit. Reason: training on large number of molecular descriptors without selection led to severe overfitting and poor AUC
Enabling Data-Driven Experimentation for High-Performance Polymer Thin Film Formulations · Georgia Tech
Lost to a baseline
Random forest cross-validation R2 (0.658) lost to simpler multiple linear regression cross-validation R2 (0.733) due to overfitting.
Lost to a baseline
Random Forest classifier test accuracy dropped more severely (from 0.78 to 0.69) compared to Support Vector Machine (stayed at 0.71) when pruned to top four LIME-selected features due to overfitting on the full feature set.
Multiscale Integration of Cross-Modal Subsurface Data for Reservoir Characterization under Label-Constrained Environments · Georgia Tech
Considered and rejected
Considered and rejected: Rejected complex non-linear regressors (Random Forest, Gradient Boosting) for speech emotion prediction because they overfit the training dataset and sacrificed cross-dataset generalizability.
Creating Links: Building an Educational Platform to Ask Relevant Questions in Education · MIT
Considered and rejected
Considered and rejected: Rejected Random Forest as the final deployed estimator due to lack of interpretability, high memory footprint, and risk of overfitting on small datasets.
Computational Intelligence for Computer-Aided Design Machine Learning Techniques for Microcontrollers Performance Screenings · IRIS - POLITO - prod
Considered and rejected
Considered and rejected: Rejected Random Forest and KNN regression for destination factor mapping in favor of Multiple Linear Regression due to severe overfitting and poor interpretability.
Leveraging the subtle : hidden factors in recommender systems · DSpace-CRIS at TU Wien
Considered and rejected
Considered and rejected: Decided against random forest and exploratory factor analysis for indicator reduction due to overfitting risks on high-dimensional data.
Considered and rejected
Considered and rejected: Decision tree learning for generalized model development (due to producing variable behavioral thresholds that were overfitted to individual runways)
Incorporation of Causal Factors Affecting Pilot Motivation for Improvement of Airport Runway and Exit Design Modeling · Virginia Tech
Tried and failed
direct execution time prediction with random forest applied to database query execution time modeling. Outcome: overfit. Reason: over-optimized slow queries at the expense of fast ones; predicting decomposed weight parameters was required
Building Instance Aware Systems using Explicit Performance Modeling · MIT
Lost to a baseline
In alternate environment validation, simpler CART and Logistic Regression models (balanced accuracy 79.7% and 78.9%) outperformed Random Forest and AdaBoost models which suffered severe overfitting.
COUNTERING SMALL UNMANNED AIRCRAFT SYSTEMS WITH ADVANCED DATA ANALYSIS AND MACHINE LEARNING · Calhoun
Tried and failed
gradient boosted trees with default hyperparameters applied to weakly supervised anomaly detection. Outcome: overfit. Reason: default tree hyperparameters were overly aggressive for weakly supervised learning labels
LOOKING FOR NEW PHYSICS: FROM DARK MATTER TO MACHINE LEARNING · Cornell
Tried and failed
gradient boosted decision tree classification applied to high-dimensional metabolomics profiles. Outcome: did not generalise. Reason: model failed to perform better than chance on an independent hold-out cohort
DEFINING THE METABOLIC LANDSCAPE OF FIBROLAMELLAR CARCINOMA · Cornell
Tried and failed
random forest on raw radar backscatter features applied to surface water extent mapping. Outcome: did not generalise. Reason: raw backscatter polarizations and incidence angle lacked sufficient discriminative power across diverse environmental conditions
Considered and rejected
Considered and rejected: Rejected Tree Parzen Estimators and Random Forests as surrogate models in BO under tight computational budgets due to over-partitioning/overfitting with few evaluations
Tried and failed
random forest hyperparameter tuning via grid search applied to time series error correction. Outcome: overfit. Reason: overemphasized lag-1 features, degrading downstream simulation accuracy and inflating variance
Tree ensembles were outperformed by gradient boosting frameworks and neural networks
Random forests and standard decision trees often lost to algorithms like XGBoost, LightGBM, neural networks, and support vector machines on complex prediction tasks. These competing architectures better captured continuous spatial gradients, complex temporal horizons, and sequential error corrections.
Considered and rejected
Considered and rejected: Rejected using Random Forest models in favor of XGBoost due to extreme computational overhead and boundary overfitting.
Tried and failed
random forest regression applied to satellite retrieval bias correction. Outcome: worse than baseline. Reason: significantly underperformed gradient boosted decision tree algorithms like LightGBM and XGBoost
Tried and failed
gradient boosted decision trees regression applied to multimodal extracted image and process features. Outcome: did not generalise. Reason: poorly captured non-linear relationships compared to neural network representations
ML-accelerated pipeline for understanding atomistic hardening of MPEAs and predicting hardness in additive manufacturing · Virginia Tech
Tried and failed
gradient boosted trees for long-horizon regression applied to renewable resource time-series forecasting. Outcome: worse than baseline. Reason: Tree ensembles struggled with complex temporal patterns over long-term forecasting horizons compared to other models.
Machine learning applications for the optimization of renewable energy systems · Iowa State
Tried and failed
gradient boosted decision trees applied to cluster membership classification. Outcome: worse than baseline. Reason: cascading weak learner errors caused poor classification performance
A network-level statistical model for friction estimation in Texas · UT Austin
Tried and failed
random forest regression for field variable surrogate applied to turbulent flow field prediction. Outcome: worse than baseline. Reason: random forest failed to capture complex spatial gradients compared to deep learning framework
Multifidelity machine learning methods for flow field prediction and aerodynamic shape optimization · Iowa State
Lost to a baseline
Random Forest on 5-stage PU UV degradation achieved only 59% accuracy with O-PTIR and 60% with ATR-FTIR, underperforming PLS-DA (80% and 51%) and SVM (82% on O-PTIR).
An investigation on the applications of advanced Infrared Spectroscopy, Spectral Imaging and Machine Learning for Polymer Characterization, including microplastics · Research Repository UCD
Lost to a baseline
Smooth Random Forest underperformed DPGBDT 1x100 on the HELOC dataset.
A consumer centred investigation of differentially private risk assessment models in consumer credit · University of Nottingham Repository
Lost to a baseline
Support Vector Machine (SVM), Random Forests, and Gaussian Processes achieved lower test dataset R2 compared to the 2-hidden-layer ANN model for predicting HFRC tensile stress-strain behavior.
Condition Assessment of Civil Infrastructure and Materials Using Deep Learning · Virginia Tech
Lost to a baseline
Single Decision Tree for HEA phase classification achieved only 67.16% mean CV accuracy, trailing Gradient Boosting (72.99%).
Accelerated Data-Driven Design of Multi-Component Chemistries for Enhanced Mechanical Performance · DSpace at SUNY Buffalo
Lost to a baseline
Gradient boosting (Model 4 R2 0.46–0.92) slightly outperformed random forest (Model 3 R2 0.33–0.98).
Considered and rejected
Considered and rejected: Rejected Decision Trees, Random Forests, Support Vector Machines, Regularized Linear Regression, and Multi-Layer Perceptrons in favor of XGBoost for Hurricast prediction due to lower performance and higher training time.
Considered and rejected
Considered and rejected: Random forest classifier rejected in favour of XGBoost due to lower test accuracy and inferior prediction calibration across SES and gender subgroups
Understanding Student Outcomes: The Role of Background, Gender, and Schools in Irish Education · Research Repository UCD
Considered and rejected
Considered and rejected: Rejected Random Forest classification in favor of Gradient Boosted Decision Trees (XGBoost) due to XGBoost's sequential error correction and higher expected performance
Models were rejected due to opacity and computational overhead compared to simpler models
Researchers rejected random forest models because their high complexity and lack of interpretability obscured individual variable effects. In addition, tree ensembles required substantially greater runtime and memory footprint without offering accuracy advantages over simpler decision trees or linear alternatives.
Lost to a baseline
Random Forest classifier consumed approximately 8x more runtime than Decision Trees despite achieving comparable accuracy, precision, and recall.
Dynamic Decomposition and Deployment of the Virtual Network Functions with Microservices using Deep Reinforcement Learning · Research Repository UCD
Considered and rejected
Considered and rejected: Rejected Random Forest classification in mlZ in favor of Decision Trees due to lack of accuracy improvement and increased complexity
Considered and rejected
Considered and rejected: Random forest regression rejected in favor of linear regression due to loss of interpretability, poor extrapolation, and inability to handle memory-bound workloads without full target-state counter inputs.
Considered and rejected
Considered and rejected: Rejected Random Forest as the final model due to its high complexity and lack of interpretability compared to Decision Trees.
Considered and rejected
Considered and rejected: Rejected Random Forest in favor of Logistic Regression due to logistic regression's superior enlisted AUC (0.836 vs 0.695), interpretability/explainability, and reduced computational complexity.
U.S. ARMY RESERVE RETENTION MODELING FOR MID-LEVEL LEADERS · Calhoun
Considered and rejected
Considered and rejected: Rejected random forest and classification tree models in favor of logistic regression because logistic regression provides clearer interpretability for individual variable effects.
PREDICTING MIDSHIPMEN'S OUTCOMES AT THE UNITED STATES NAVAL ACADEMY · Calhoun
Considered and rejected
Considered and rejected: Random forest regression in favor of multiple linear regression due to similar performance and greater simplicity/interpretability of MLR.
Considered and rejected
Considered and rejected: Rejected using non-linear and random forest models to focus solely on linear logistic regression of binary outcomes.
Considered and rejected
Considered and rejected: Tree-based models (such as XGBoost and Random Forests) and Support Vector Machines were rejected in favor of LASSO regression due to multicollinearity issues, extensive hyperparameter tuning requirements, and lack of automatic feature elimination
Tumor Immunogenicity Unlocked: Multi-omics Models Predict Immunotherapy Response and Survival · Publikationssystem UB Tuebingen
Considered and rejected
Considered and rejected: Rejected machine learning (neural networks/random forests) due to high false-positive rates and uninterpretable feature explanations.
Development of computational tools for variant calling in single-cell RNAseq · Oxford
Tree models suffered from poor recall and erratic behavior under class imbalance
When applied to imbalanced target distributions, tree ensembles exhibited poor sensitivity and scattered, noisy predictions. Automated feature elimination and resampling strategies often discarded weak signals essential for detecting rare minority classes.
Lost to a baseline
Random forest demographic-only model yielded an artifactually high sensitivity of 0.308 due to rare outcome imbalances, performing erratically compared to logistic regression baselines.
PREDICTING SUICIDE: UTILITY OF SOCIAL DETERMINANTS OF HEALTH AND ACCESS DATA · JScholarship
Lost to a baseline
Non-parametric models (Random Forest and XGBoost) were outperformed by simpler parametric models (Logistic Regression and LDA) in sensitivity and balanced accuracy when using synthetic resampling (e.g., LR achieved 0.60 sensitivity vs RF 0.32 and XGBoost 0.36 under SMOTE).
Evaluating Factors Contributing to Crash Severity Among Older Drivers: Statistical Modeling and Machine Learning Approaches · Virginia Tech
Lost to a baseline
Logit model achieved higher specificity (0.876) on unbalanced rCSI test data than Random Forest (0.348) and XGBoost (0.627)
Essays in Applied Economics Linking Policy, Obesity, and Health Economics · Texas Tech
Tried and failed
gradient boosted decision trees classification applied to imbalanced fine-grained time-series state prediction. Reason: severe class imbalance produced noisy, scattered predictions with low precision and recall
Understanding and Predicting Sit-Stand Desk Usage Patterns and Willingness among Knowledge Workers: A Data-Driven Approach · Virginia Tech
Tried and failed
recursive feature elimination with cross validation applied to gradient boosted trees on imbalanced data. Outcome: worse than baseline. Reason: feature selection discarded weakly predictive features crucial for detecting the rare minority class
Lost to a baseline
Traditional Random Forest and Gradient-Boosted Trees suffered from lower recall (0.304 and 0.607, respectively) on HCM classification with CMR features compared to SVM-RBF (0.857) due to class imbalance.
Considered and rejected
Considered and rejected: Rejected machine learning approaches (decision trees and random forest) for sensor-drone assignment due to poor constraint handling, class imbalance, and lack of generalizability
Development of a Natural Disaster Information System for RPAS Mission Planning and Monitoring · Queens University Institutional Repository
Feature importance metrics and hyperparameter tuning methods failed or introduced bias
Impurity-based feature importance exhibited systematic bias toward continuous variables and features with many categories. Furthermore, hyperparameter tuning via grid search or genetic algorithms often failed to beat untuned base models or removed predictive feature shortcuts.
Tried and failed
random forest mean decrease in impurity applied to feature importance estimation. Reason: biased toward continuous variables and features with many categories
Tried and failed
hyperparameter grid search for random forest applied to tabular demographic and electoral feature classification. Outcome: worse than baseline. Reason: regularization removed single-feature shortcuts that had driven unconstrained performance
Considered and rejected
Considered and rejected: Rejected evaluating genetic algorithm fitness using Random Forest instead of Logistic Regression because Random Forest's inherent feature importance mechanisms mitigate redundancy and obscure individual feature contributions during GA optimization.
Non-Invasive cancer detection: computational applications in liquid biopsy and radiomics · IRIS - UNITN - prod
Tried and failed
untuned random forest with few estimators applied to blood biomarker disease classification. Outcome: worse than baseline. Reason: insufficient ensemble size and lack of hyperparameter tuning caused extremely poor classification performance
Machine-Learning Aided Diagnosis of Alzheimer's Disease · UT Austin
Lost to a baseline
Feature selection on gas-phase adiabatic ionization potential using random forest failed to improve performance and slightly worsened ionization potential predictions relative to full feature sets.
Tried and failed
random forest Bayesian optimization for configuration tuning applied to runtime configuration optimization. Outcome: worse than baseline. Reason: direct flag-to-runtime mapping struggled to navigate high-dimensional configuration spaces effectively compared to random search
Accelerating regression testing through test environment tuning · UT Austin
Lost to a baseline
In Experiment 2, Grid Search and Genetic Algorithm tuning of Random Forest (MSE 0.0061, R2 0.8086) lost to the base untuned Random Forest (MSE 0.0042, R2 0.8158).
Using Data Analytics and Machine Learning in Sustainable Forest Management from Remote Sensing Data · YorkSpace
Output variance compression caused inaccurate probability and extreme value predictions
Averaging mechanisms across trees compressed prediction variance, leading models to systematically over-predict low values and under-predict high values. As a result, probability estimates degraded near distribution boundaries and failed to capture extreme tail exceedances.
Tried and failed
random forest regression on continuous clinical scores applied to blood biomarker disease severity prediction. Reason: compressed output variance caused regression to fail at extreme values, resulting in high false positives
Machine-Learning Aided Diagnosis Of Alzheimer's Disease · UT Austin
Tried and failed
indirect mapping predicting multi-item components via random forests applied to health-related utility score estimation. Reason: systematically under-predicted low scores and over-predicted high scores, inflating the overall predicted mean
Tried and failed
random forests with tail oversampling and residual fitting applied to extreme value exceedance prediction. Reason: techniques were ineffective at resolving tail smoothing in predictions
Data-Driven Methods for Modeling Emissions and Atmospheric Composition · Harvard
Considered and rejected
Considered and rejected: Decided against Random Forests for surrogate modeling because of large prediction errors on sampled points and lack of reliable predictive variance estimates.
Accelerating HLS Autotuning of Large, Highly-parameterized Reconfigurable SoC Mappings · Penn
Tried and failed
random forest probability estimation applied to molecular synthetic accessibility prediction. Reason: uncalibrated models yielded poor probability estimates near extreme thresholds
Discovery of synthesisable organic materials · Imperial
Left open by the authors
Problems the authors named and did not get to.
Left open
Perform a grid search over hyperparameter space instead of random search for the Random Forest and XGBoost Alzheimer's diagnosis models. Blocker: Access to the underlying clinical biomarker dataset used in the thesis.
Machine-Learning Aided Diagnosis Of Alzheimer's Disease · UT Austin
Left open
Evaluate varying decision tree depths and random forest models for embedding subspace mapping in neural network generalization prediction. Blocker: None
DEPENDABLE NEURAL NETWORKS FOR SAFETY CRITICAL TASKS · JScholarship
Left open
Prospectively test decision tree and random forest models across diverse healthcare systems to evaluate clinical efficacy in childhood genetic epilepsy. Blocker: Requires prospective clinical deployment and access to private electronic medical record data across multiple healthcare systems.
Quantitative Informatics Approaches to Characterize and Predict Childhood Genetic Epilepsies · Penn
Left open
Evaluate statistical and systematic uncertainty behaviors under Random Forest, SVM, and deep neural network models on student response coding datasets. Blocker: Access to private student physics response datasets used in the thesis.
Evaluating language models applied to student thinking about experiments · Cornell
Left open
Benchmark healthcare capital project performance by evaluating Random Forests, KNN, and DEA grouping algorithms on larger datasets. Blocker: Requires larger proprietary or restricted healthcare capital project BIM and performance datasets
Novel approaches to benchmark capital project performance : an application to healthcare projects · UT Austin
Left open
Develop online prediction for the random forest framework in aerodynamic shape optimization and test alternative architectures like GNN, RNN, and GAN. Blocker: Lack of specific evaluation metrics, clear baseline integration details, or designated datasets for the proposed alternative ML architectures.
Multifidelity machine learning methods for flow field prediction and aerodynamic shape optimization · Iowa State
Left open
Implement decision trees or random forests to combine covert channel detection statistical tests instead of logistic regression. Blocker: None
Real-Time Detection of Storage Covert Channels · Carleton University Institutional Repository
Left open
Train and compare alternative machine learning regression models against random forest for estimating the truncation parameter of truncated exponential distributions. Blocker: None
Estimation of Parameters for Truncated Exponential Distribution · unevada
Left open
Evaluate the reduced feature subsets produced by the fuzzy feature selection methods using classifiers like SVM and Random Forest. Blocker: None
An investigation of fuzzy methods and meta learning for feature selection · University of Nottingham Repository
Left open
Evaluate alternative surrogate models and acquisition functions beyond Random Forest and Expected Improvement for tuning LSM trees in Onix. Blocker: None
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.