Nonparametric Machine Learning Research Articles

Analyzing and forecasting air pollution concentration in the capital and Southern Thailand using a lag-dependent Gaussian process model.

The air pollution problem has now amassed worldwide attention due to its multifaceted harm to human health. Exploring the concentration of air pollution and improving forecast have important consideration worldwide. In this research, we analyze the air pollution concentration of Southern Thailand and compare it with the central region. Also, we proposed a methodology based on the lag-dependent Gaussian process (LDGP), a Bayesian non-parametric machine learning model, with a stable optimization approach, which is a cluster-based multi-starter technique based on the Nelder-Mead optimizer. This model also provides the confidence band for forecasted values. We also used autoregressive deep neural network (AR-DNN), autoregressive random forest (AR-RF), gradient boosting (GB), and K-nearest neighbors (KNN) models. A comparison of the proposed methodology was performed on the daily air pollution data collected from the southern provinces and also from the capital of Thailand from 1 January 2018 to 31 December 2022. We used well-established performance evaluation measures to compare the performance of the models. To evaluate the bias due to overfit, we performed a tenfold cross-validation for all the pollutants in each region and compared the models to choose the best one. Moreover, we explored the concentration of air pollution in these regions. Results of descriptive analysis revealed that Bangkok had a much higher concentration of air pollution as compared to the southern region. However, the southern region had higher exposure to PM air pollutants as per WHO recommendations and also had higher exposure to O3 and CO levels. The proposed LDGP model outperformed the other machine learning models for forecasting all air pollutants. Hence, it is recommended to be used by experts for further research and studies with different kernel functions. This research is also expected to contribute to local government planning and prevention and worldwide use of the same methodology for the sustainability of public health.

Development of a quantitative prediction algorithm for human cord blood-derived CD34+ hematopoietic stem-progenitor cells using parametric and non-parametric machine learning models

The transplantation of CD34+ hematopoietic stem-progenitor cells (HSPCs) derived from cord blood serves as the standard treatment for selected hematological, oncological, metabolic, and immunodeficiency disorders, of which the dose is pivotal to the clinical outcome. Based on numerous maternal and neonatal parameters, we evaluated the predictive power of mathematical pipelines to the proportion of CD34+ cells in the final cryopreserved cord blood product adopting both parametric and non-parametric algorithms. Twenty-four predictor variables associated with the cord blood processing of 802 processed cord blood units randomly sampled in 2020–2022 were retrieved and analyzed. Prediction models were developed by adopting the parametric (multivariate linear regression) and non-parametric (random forest and back propagation neural network) statistical models to investigate the data patterns for determining the single outcome (i.e., the proportion of CD34+ cells). The multivariate linear regression model produced the lowest root-mean-square deviation (0.0982). However, the model created by the back propagation neural network produced the highest median absolute deviation (0.0689) and predictive power (56.99%) in comparison to the random forest and multivariate linear regression. The predictive model depending on a combination of continuous and discrete maternal with neonatal parameters associated with cord blood processing can predict the CD34+ dose in the final product for clinical utilization. The back propagation neural network algorithm produces a model with the highest predictive power which can be widely applied to assisting cell banks for optimal cord blood unit selection to ensure the highest chance of transplantation success.

GEOGRAPHICALLY WEIGHTED MACHINE LEARNING MODEL FOR ADDRESSING SPATIAL HETEROGENEITY OF PUBLIC HEALTH DEVELOPMENT INDEX IN JAVA ISLAND

Random Forest (RF) machine learning models have emerged as a prominent algorithm, addressing problems arising from the sole use of decision trees, such as overfitting and instability. However, conventional RF has global coverage that may need to capture spatial variations better. Based on the analysis of the level of public health development, the relationship between the level of health development and risk factors can vary spatially. We use a modified RF algorithm called Geographically Weighted Random Forest (GW-RF) to address this challenge. GW-RF, as a tree-based non-parametric machine learning model, can help explore and visualize relationships between the Public Health Development Index (PHDI) as response variables and factors that are indicators at the district level. GW-RF output is compared with global output, which is RF in 2018 using the percentage of the population with access to clean/decent water (X1), consumption of eggs and milk per capita per week (X2), number of healthcare facilities per 1000 people (X3), number of doctors per 1000 people (X4), pure participation rate ratio female/male (X5), percentage of households that have hand washing facilities with soap and water (X6) as independent variables. Our results show that the non-parametric GW-RF model shows high potential for explaining spatial heterogeneity and predicting PHDI versus a global model when including six major risk factors. However, some of these predictions mean little. Findings of spatial heterogeneity using GW-RF show the need to consider local factors in approaches to increasing PHDI values. Spatial analysis of PHDI provides valuable information for determining geographic targets for areas whose PHDI values need to be improved.

Soil sampling design matters - Enhancing the efficiency of digital soil mapping at the field scale

Optimisation of sampling design (methods chosen to select the samples) and sample size (number of samples) remains a key challenge in digital soil mapping, especially in the area of precision farming with the expected economic benefits from the introduction of new technologies. As the existing information is available in the form of relevant environmental covariates, its combination with non-parametric machine learning techniques requires careful planning from the initial field sampling to the final production of digital soil maps. The aim of this study is to compare widely used covariate-wise sampling designs combined with variable sample sizes for supervised prediction of common soil drivers of agricultural productivity (pH, soil organic carbon, soil macronutrients) in a real case study of a field (35 ha) with heterogeneous soil properties. From a total of 200 samples, we evaluated different sample sets where 10, 30 and 60 field samples were selected by conditioned Latin Hypercube Sampling (cLHS) and Feature Space Coverage Sampling (FSCS) to calibrate random forest (RF) models. The evaluation was performed on independently in-situ sampled test points. In addition to these datasets, we also compared the investigated methods with Simple Random Sampling (SRS) in a numerical benchmark experiment with increasing sample size, comparing the global accuracies of the predicted maps on the test points, but using interpolated maps as the artificial true population for each soil characteristic. The results of the study in both the field experiment and the numerical experiment showed slightly better results for the FSCS method, especially when the number of samples was small. At smaller training sample sizes, the risk of insufficiently accurate prediction models was slightly lower for FSCS and the difference decreased as the sample size increased. Nevertheless, sample size proved to be the most important factor in the accuracy of RF models, regardless of the sampling technique. The results suggest that a sample size between 18 and 30 training samples (0.6 to 1 sample ha−1) seems plausible for covariate-wise predictions using RF at field scale in our case study. The relative importance of each auxiliary variable for each RF calibration was also assessed for the field experiment. The results showed that the additional introduction of spatial proxies overshadowed the importance of other covariates, but only significantly improved the model calibration at larger sample sizes. The calibrated models without spatial proxies showed the strongest effect of remotely sensed surface characteristics.

Construction of Al–Si interatomic potential based on Bayesian active learning

Nanoscale simulations for optimizing the performance and processing of Al–Si alloys are currently facing two major obstacles: the scarcity of high-quality semi-empirical potentials tailored to complex alloy systems, and the prohibitively high computational cost associated with ab initio molecular dynamics simulations. In order to enhance simulation efficiency and accuracy of the Al–Si alloys’ microstructural evolution, this study employs a dynamic active learning technique, FLARE, to develop a non-parametric machine learning potential that combines the high accuracy of density functional theory (DFT) with the efficiency of classical molecular dynamics (MD). Without relying on extensive initial ab initio molecular dynamics data or existing databases, collection of the necessary data is progressively made during the active learning process, thereby constructing a potential capable of accurately simulating the structure and dynamics of high-temperature Al–Si alloys. By comparing with experimental measurements and ab initio molecular dynamics calculations, the high accuracy and computational efficiency of this potential is demonstrated in predicting energy, force, structure, and dynamic properties. The results provide novel theoretical insights for optimizing Al–Si alloy processing and underscore the usefulness of active learning methods in constructing high-accuracy potentials.

Refined semi-empirical models for soil moisture retrieval in spaceborne GNSS-Reflectometry: Evaluation across diverse land cover types

The satellite-based Global Navigation Satellite System-Reflectometry (GNSS-R) has emerged as a promising technique for detecting surface soil moisture (SSM), which plays a pivotal role in various applications. Different approaches have been developed for SSM retrieval, including parametric semi-empirical models and non-parametric machine learning methods. This study, however, specifically focuses on constructing and evaluating semi-empirical fit models. To this end, the study compares the effectiveness of our enhanced pixel-by-pixel model, the land cover-based linear model, and the Reflectivity-Vegetation-Roughness (R-V-R) model across different land surface types, aiming to both evaluate their efficacy and identify potential limitations. In the assessment, various factors were taken into consideration, such as the correction for vegetation and roughness attenuation, fitting functions employed, and the utilization of a lookup table (LUT). The results of the evaluation showed variations in the performance of the retrieval models across different land cover types, highlighting the impact of the choice of fitting functions and attenuation correction strategies on the accuracy of soil moisture retrieval. The pixel-by-pixel model demonstrated the highest prediction accuracy, with an unbiased root mean square difference (ubRMSD) of 0.056 cm3/cm3 and a correlation coefficient of 0.896. By showcasing these outcomes, this research underscores the significance of accounting for surface conditions and integrating relevant data to enhance the accuracy of GNSS-R SSM retrieval, thereby contributing to the advancement of SSM monitoring methodologies.

Groundwater potential mapping in arid and semi-arid regions of kurdistan region of Iraq: A geoinformatics-based machine learning approach

Groundwater (GW) is a crucial and increasingly scarce natural resource, that is affected by climate change and mismanagement. To manage GW resources effectively, it is crucial to accurately identify GW potential zones (GWPZs) using modern techniques. This study aimed to employ and assess geoinformatics-based machine learning (ML) models to delineate GWPZs in the Akre district, Kurdistan region of Iraq. Six nonparametric ML models were used: a support vector machine (SVM), k-nearest neighbours (KNN), decision tree (DT), random forest (RF), gradient boost DT (GBDT), and extreme gradient boosting (XGBoost). These models were trained on diverse GWPZ-favourable influencing factors, encompassing topographic, hydrological, geological, and environmental aspects. The findings of this study revealed that the XGBoost model outperformed the other nonparametric models in terms of best-fit performance and accuracy in generating a GW potential map (GWPM), achieving a R2 of 0.88, a root mean square error (RMSE) of 11.348, and a mean absolute error (MAE) of 6.623. Notably, over half of the study area (53%) was categorised as having high or very high GWPZs, primarily in the low-lying Rovia Plain. The study identified rainfall, elevation, lineament density (LD), drainage density (Dd), topographic wetness index (TWI), and slope as the most significant factors influencing GWPZ modelling. This study provides a comprehensive framework for GW resource management, ecological conservation, and urban development planning. These insights are crucial for stakeholders, policymakers, and local authorities in strategic resource planning and environmental stewardship.

Risk assessment of bridge construction investigated using random forest algorithm

Bridge construction collapse is one of the most common bridge safety accidents. At present, evaluation results are often affected by the ability and experience of the assessor. Therefore, it is difficult to quickly, accurately and effectively evaluate the risk in the process of bridge construction. Moreover, key factors that can prevent accidents can hardly find from the existing bridge construction safety management and evaluation method. This paper analyzes and classifies the artificial and environmental risk factors that affect the bridge construction stage, and establishes 26 risk factors in 5 categories according to the characteristics of bridge construction and the actual situation of the project. Random forest (RF) algorithm is a non-parametric machine learning method based on decision tree, which does not need to be scored by experts in advance and avoids the influence of subjective factors. Compared with other analysis methods, random forest algorithm has the advantages of accurate and robust risk assessment results. Based on the advantages of random forest algorithm and the characteristics of bridge construction risk, this paper uses random forest algorithm to evaluate the bridge construction risk, and ranks the importance of indicators, and identify the index that has a greater influence on the risk. In order to verify the applicability and feasibility of the proposed method, a typical urban complex pedestrian bridge was taken as an example for actual engineering evaluation and verification. The results obtained are basically consistent with the actual risk assessment results of the pedestrian bridge.

Outcome risk model development for heterogeneity of treatment effect analyses: a comparison of non-parametric machine learning methods and semi-parametric statistical methods

BackgroundIn randomized clinical trials, treatment effects may vary, and this possibility is referred to as heterogeneity of treatment effect (HTE). One way to quantify HTE is to partition participants into subgroups based on individual’s risk of experiencing an outcome, then measuring treatment effect by subgroup. Given the limited availability of externally validated outcome risk prediction models, internal models (created using the same dataset in which heterogeneity of treatment analyses also will be performed) are commonly developed for subgroup identification. We aim to compare different methods for generating internally developed outcome risk prediction models for subject partitioning in HTE analysis.MethodsThree approaches were selected for generating subgroups for the 2,441 participants from the United States enrolled in the ASPirin in Reducing Events in the Elderly (ASPREE) randomized controlled trial. An extant proportional hazards-based outcomes predictive risk model developed on the overall ASPREE cohort of 19,114 participants was identified and was used to partition United States’ participants by risk of experiencing a composite outcome of death, dementia, or persistent physical disability. Next, two supervised non-parametric machine learning outcome classifiers, decision trees and random forests, were used to develop multivariable risk prediction models and partition participants into subgroups with varied risks of experiencing the composite outcome. Then, we assessed how the partitioning from the proportional hazard model compared to those generated by the machine learning models in an HTE analysis of the 5-year absolute risk reduction (ARR) and hazard ratio for aspirin vs. placebo in each subgroup. Cochran’s Q test was used to detect if ARR varied significantly by subgroup.ResultsThe proportional hazard model was used to generate 5 subgroups using the quintiles of the estimated risk scores; the decision tree model was used to generate 6 subgroups (6 automatically determined tree leaves); and the random forest model was used to generate 5 subgroups using the quintiles of the prediction probability as risk scores. Using the semi-parametric proportional hazards model, the ARR at 5 years was 15.1% (95% CI 4.0–26.3%) for participants with the highest 20% of predicted risk. Using the random forest model, the ARR at 5 years was 13.7% (95% CI 3.1–24.4%) for participants with the highest 20% of predicted risk. The highest outcome risk group in the decision tree model also exhibited a risk reduction, but the confidence interval was wider (5-year ARR = 17.0%, 95% CI= -5.4–39.4%). Cochran’s Q test indicated ARR varied significantly only by subgroups created using the proportional hazards model. The hazard ratio for aspirin vs. placebo therapy did not significantly vary by subgroup in any of the models. The highest risk groups for the proportional hazards model and random forest model contained 230 participants each, while the highest risk group in the decision tree model contained 41 participants.ConclusionsThe choice of technique for internally developed models for outcome risk subgroups influences HTE analyses. The rationale for the use of a particular subgroup determination model in HTE analyses needs to be explicitly defined based on desired levels of explainability (with features importance), uncertainty of prediction, chances of overfitting, and assumptions regarding the underlying data structure. Replication of these analyses using data from other mid-size clinical trials may help to establish guidance for selecting an outcomes risk prediction modelling technique for HTE analyses.

Can Multi-Temporal Vegetation Indices and Machine Learning Algorithms Be Used for Estimation of Groundnut Canopy State Variables?

The objective of this research was to assess the feasibility of remote sensing (RS) technology, specifically an unmanned aerial system (UAS), to estimate Bambara groundnut canopy state variables including leaf area index (LAI), canopy chlorophyll content (CCC), aboveground biomass (AGB), and fractional vegetation cover (FVC). RS and ground data were acquired during Malaysia’s 2018/2019 Bambara groundnut growing season at six phenological stages; vegetative, flowering, podding, podfilling, maturity, and senescence. Five vegetation indices (VIs) were determined from the RS data, resulting in single-stage VIs and cumulative VIs (∑VIs). Pearson’s correlation was used to investigate the relationship between canopy state variables and single stage VIs and ∑VIs over several stages. Linear parametric and non-linear non-parametric machine learning (ML) regressions including CatBoost Regressor (CBR), Random Forest Regressor (RFR), AdaBoost Regressor (ABR), Huber Regressor (HR), Multiple Linear Regressor (MLR), Theil-Sen Regressor (TSR), Partial Least Squares Regressor (PLSR), and Ridge Regressor (RR) were used to estimate canopy state variables using VIs/∑VIs as input. The best single-stage correlations between canopy state variables and VIs were observed at flowering (r > 0.50). Moreover, ∑VIs acquired from vegetative to senescence stage had the strongest correlation with all measured canopy state variables (r > 0.70). In estimating AGB, MLR achieved the best testing performance (R2 = 0.77, RMSE = 0.30). For CCC, RFR excelled with R2 of 0.85 and RMSE of 2.88. Most models performed well in FVC estimation with testing R2 of 0.98–0.99 and low RMSE. For LAI, MLR stood out in testing with R2 of 0.74, and RMSE of 0.63. Results demonstrate the UAS-based RS technology potential for estimating Bambara groundnut canopy variables.

Interpretable data-driven prediction methods for atmospheric chloride deposition rate

Precise estimation of atmospheric chloride deposition is crucial in environmental science, providing valuable insights into atmospheric pollution, metal corrosion, and the hydrologic cycle. The distribution of atmospheric chloride deposition involves multiple processes of production, transportation, and deposition, making it a problem with well-defined physical mechanisms. Although numerous parametric models have been proposed for atmospheric chloride distribution prediction, they present challenges in terms of generalizability and prediction accuracy due to simplifications of the holistic process. In contrast, non-parametric machine learning algorithms can analyze complex issues and have good prediction performance, but they are used less often in atmospheric chloride deposition modeling. Therefore, we develop a genetic algorithm optimized quantile regression forest (GA-QRF) model, incorporating environmental variables associated with the holistic deposition process to predict the distribution of atmospheric chloride deposition. QRF is an ensemble learning method known for delivering precise and interpretable predictions, coupled with effective uncertainty estimation. The proposed model results achieved an R2 value of 0.935 for median prediction and 98.2% coverage of the intervals for uncertainty quantification on a dataset collected from six countries. In addition, in order to verify the consistency of the black-box machine learning model with the domain knowledge, several interpretation methods are used to deconstruct the proposed model. Considering the spatial variability of atmospheric chloride deposition, the proposed model was further used to create deposition distribution maps.

Assessing above-ground biomass in reforested urban landscapes using machine learning and remotely sensed data

ABSTRACT Urban reforestation mitigates climate change by sequestering carbon, but quantifying carbon gains requires accurate aboveground biomass estimation. This study estimated carbon sequestration in a reforested urban landscape using PlanetScope, Sentinel-1A, Sentinel-2A, SRTM data, and field measurements. Non-parametric machine learning algorithms (k-nearest neighbor, support vector machines, extreme gradient boosting, random forests) with 39 predictor features generated aboveground biomass density maps. The extreme gradient boosting model performed best, predicting 4.1–286.5t ha-1 aboveground biomass, demonstrating its effectiveness for modeling reforested biomass with multi-source data. Findings highlight extreme gradient boosting’s promise for urban biomass estimation, the importance of multi-source data, and machine learning’s potential in addressing environmental challenges like climate change.

Machine learning enables electrical resistivity modeling of printed lines in aerosol jet 3D printing

Among various non-contact direct ink writing techniques, aerosol jet printing (AJP) stands out due to its distinct advantages, including a more adaptable working distance (2–5 mm) and higher resolution (~ 10 μm). These characteristics make AJP a promising technology for the precise customization of intricate electrical functional devices. However, complex interactions among the machine, process, and materials result in low controllability over the electrical performance of printed lines. This significantly affects the functionality of printed components, thereby limiting the broad applications of AJP. Therefore, a systematic machine learning approach that integrates experimental design, geometrical features extraction, and non-parametric modeling is proposed to achieve printing quality optimization and electrical resistivity prediction for the printed lines in AJP. Specifically, three classical convolutional neural networks (CNNs) architectures are compared for extracting representative features of printed lines, and an optimal operating window is identified to effectively discriminate better line morphology from inferior printed line patterns within the design space. Subsequently, three representative non-parametric machine learning techniques are employed for resistivity modeling. Following that, the modeling performances of the adopted machine learning methods were systematically compared based on four conventional evaluation metrics. Together, these aspects contribute to optimizing the printed line morphology, while simultaneously identifying the optimal resistivity model for accurate predictions in AJP.

Data-driven atmospheric corrosion prediction model for alloys based on a two-stage machine learning approach

Alloy material corrosion is a comprehensive problem affected by multi-dimensional factors, including the time, environment, and material element. Its corrosion prediction has been facing the challenges of complex mechanism and small sample size. However, parametric methods are limited in terms of modeling accuracy and generalization performance for high-dimensional feature problems. Non-parametric machine learning algorithms, while having excellent modeling capabilities, need to be optimized in conjunction with the structure of the corrosion problem. Therefore, we develop a two-stage hybrid intelligent machine learning model for improving alloy corrosion prediction accuracy. The model takes the time, environment, and material information of the alloy corrosion process as inputs and the corrosion rate as output. The first stage of the model utilizes tree ensemble learning (TEL) models for initial prediction modeling and feature importance knowledge mining. In the second stage, the neural network (NN) is used to enhance the fusion of the multi-model prediction outputs, and it is optimized using feature importance modification (FIM). The proposed TEL-FIMNN model shows considerable performance improvement over the TEL and ANN models on real low-alloy steel corrosion dataset. Compared to other advanced multi-model fusion methods such as linear regression (LR), support vector regression (SVR), random forest (RF), and Bayesian model averaging (BMA), the proposed model also shows superior results. The TEL-FIMNN model provide a new structure from the perspectives of knowledge mining and fusion enhancement, which has good potential for multi-dimensional corrosion prediction.

Non-Parametric Machine Learning Modeling of Tree-Caused Power Outage Risk to Overhead Distribution Powerlines

Trees in proximity to power lines can cause significant damage to utility infrastructure during storms, leading to substantial economic and societal costs. This study investigated the effectiveness of non-parametric machine learning algorithms in modeling tree-related outage risks to distribution power lines at a finer spatial scale. We used a vegetation risk model (VRM) comprising 15 predictor variables derived from roadside tree data, landscape information, vegetation management records, and utility infrastructure data. We evaluated the VRM’s performance using decision tree (DT), random forest (RF), k-Nearest Neighbor (k-NN), extreme gradient boosting (XGBoost), and support vector machine (SVM) techniques. The RF algorithm demonstrated the highest performance with an accuracy of 0.753, an AUC-ROC of 0.746, precision of 0.671, and an F1-score of 0.693. The SVM achieved the highest recall value of 0.727. Based on the overall performance, the RF emerged as the best machine learning algorithm, whereas the DT was the least suitable. The DT reported the lowest run times for both hyperparameter optimization (3.93 s) and model evaluation (0.41 s). XGBoost and the SVM exhibited the highest run times for hyperparameter tuning (9438.54 s) and model evaluation (112 s), respectively. The findings of this study are valuable for enhancing the resilience and reliability of the electric grid.

Improving Amazon EC2 Spot Instances Price Prediction using Machine Learning Algorithm

Spot instances were introduced by Amazon EC2 in December 2009 to sell its spare capacity through auction based market mechanism. Despite its extremely low prices, cloud spot market has low utilization. Spot pricing being dynamic, spot instances are prone to out-of bid failure. Bidding complexity is another reason why users today still fear using spot instances. This work aims to present Regression Random Forests (RRFs) model to predict one-week-ahead and one-day-ahead spot prices. The prediction would assist cloud users to plan in advance when to acquire spot instances, estimate execution costs, and also assist them in bid decision making to minimize execution costs and out-of-bid failure probability. Simulations with 12 months real Amazon EC2 spot history traces to forecast future spot prices show the effectiveness of the proposed technique. Comparison of RRFs based spot price forecasts with existing non-parametric machine learning models reveal that RRFs based forecast accuracy outperforms other models. We measure predictive accuracy using MAPE, MCPE, OOBError and speed. Evaluation results show that

Unveiling the effect of composition on nuclear waste immobilization glasses’ durability by nonparametric machine learning

Ensuring the long-term chemical durability of glasses is critical for nuclear waste immobilization operations. Durable glasses usually undergo qualification for disposal based on their response to standardized tests such as the product consistency test or the vapor hydration test (VHT). The VHT uses elevated temperature and water vapor to accelerate glass alteration and the formation of secondary phases. Understanding the relationship between glass composition and VHT response is of fundamental and practical interest. However, this relationship is complex, non-linear, and sometimes fairly variable, posing challenges in identifying the distinct effect of individual oxides on VHT response. Here, we leverage a dataset comprising 654 Hanford low-activity waste (LAW) glasses across a wide compositional envelope and employ various machine learning techniques to explore this relationship. We find that Gaussian process regression (GPR), a nonparametric regression method, yields the highest predictive accuracy. By utilizing the trained model, we discern the influence of each oxide on the glasses’ VHT response. Moreover, we discuss the trade-off between underfitting and overfitting for extrapolating the material performance in the context of sparse and heterogeneous datasets.

Late Life Cognitive Function Trajectory Among the Chinese Oldest-Old Population—A Machine Learning Approach

ABSTRACT Informed by the biopsychosocial framework, our study uses the Chinese Longitudinal Healthy Longevity Survey (CLHLS) dataset to examine cognitive function trajectories among the oldest-old (80+). Employing K-means clustering, we identified two latent groups: High Stability (HS) and Low Stability (LS). The HS group maintained satisfactory cognitive function, while the LS group exhibited consistently low function. Lasso regression revealed predictive factors, including socioeconomic status, biological conditions, mental health, lifestyle, psychological, and behavioral factors. This data-driven approach sheds light on cognitive aging patterns and informs policies for healthy aging. Our study pioneers non-parametric machine learning methods in this context.

Ground-Motion Model for Small-to-Moderate Potentially Induced Earthquakes Using an Ensemble Machine Learning Approach for CENA

ABSTRACT Ground-motion models (GMMs) are vital in assessing probabilistic seismic hazards and uncertainties. This study develops new GMMs benefiting from nonparametric machine learning algorithms, including artificial neural networks, kernel ridge, random forest, and gradient boosting regression techniques for small-to-moderate potentially induced earthquakes in central and eastern North America (CENA). As part of this study, we evaluate the performance of different machine learning models in estimating peak ground acceleration (PGA) and 17 spectral accelerations based on the moment magnitude (Mw), hypocentral distance (Rhypo), and the timed-average shear-wave velocity of the upper 30 m of soil (VS30). To train the algorithms, we have utilized a database of nearly 31,000 ground motions with small and moderate moment magnitudes ranging from 3.0 to 5.8, recorded within a hypocentral distance of less than 200 km in CENA. Typically, for GMM development, analysts employ linear regression-based models with predefined functional forms. The requirement for predefined functional forms can restrict the use of complicated and nonlinear equations to improve performance. Although the conventional regression model is more interpretable, machine learning can achieve a better result given sufficient training data. The results of error metrics reveal that gradient-boosting regression provides a better performance. Furthermore, a machine learning ensemble method is used to combine the regression results of four machine learning algorithms. The ensemble method improves the GMM performance and provides smoother results.

318 Discovering Subgroups with Supervised Machine Learning Models for Heterogeneity of Treatment Effect Analysis

OBJECTIVES/GOALS: The goal of the study is to provide insights into the use of machine learning methods as a means to predict heterogeneity of treatment effect (HTE) in participants of randomized clinical trials. METHODS/STUDY POPULATION: Using data from 2,441 participants enrolled in the ASPirin in Reducing Events in the Elderly (ASPREE) randomized controlled trial of daily low-dose aspirin vs placebo in the United States, we developed multivariable risk prediction models for the composite outcome of dementia, disability, or death. We used two machine learning techniques, decision trees and random forests, to develop novel non-parametric outcomes classifiers and generate risk-based subgroups. The comparator method was an extant semi-parametric proportional hazards predictive risk model. We then assessed HTE by examining the 5-year absolute risk reduction (ARR) of aspirin vs placebo in each risk subgroup. RESULTS/ANTICIPATED RESULTS: In the random forest classifier, the ARR at 5 years in the highest risk quintile was 13.7% (95% CI 3.1% to 24.4%). For the semi-parametric proportional hazards model, the ARR in the highest risk quintile was 15.1% (95% CI 4.0% to 26.3%). These results were comparable and provide evidence of the viability of internally developed parsimonious non-parametric machine learning models for HTE analysis. The decision tree model results (5-year ARR = 17.0%, 95% CI= -5.4% to 39.4% in the highest risk subgroup) exhibited more uncertainty in the results. DISCUSSION/SIGNIFICANCE: None of the models detected significant HTE on the relative scale; there was substantial HTE on the absolute scale in three of the models. Treatment benefit on the absolute scale may be regarded as bearing greater clinical importance and may be present even in the absence of benefit on the relative scale.