Mean Decrease Accuracy Research Articles

Parameter importance analysis: Random forest approach

Abstract During surface roughness modelling, it is crucial to determine the parameters with the highest predictive power since these are the outcome drivers. Based on out-of-bag (OOB) mean square error, the following Random Forest techniques have been utilized to determine parameter importance: mean decrease in accuracy and total increase in node purity. Validation of the results has been achieved using the Bayesian linear regression technique. The PMMA machining experiment has been designed by the Central Composite Design (CCD) Face Centered technique. Cutting speed, feed rate and depth of cut are the control parameters, while surface quality is the dependent parameter. The authors established that the random forest regression algorithm yields an OOB mean squared error of 0.113 and that the OOB mean squared error decreases with increasing number of trees for validation dataset. On the other hand, the OOB mean squared error increases with increasing number of trees for training dataset. Both the mean decrease in accuracy and total increase in node purity techniques reveal that the order of decreasing machining parameter importance is as follows: cutting speed, depth of cut and feed rate. Validation of the obtained results yields the same outcome. Hence, feed rate may be omitted from models for faster and simpler surface roughness prediction.

Effects of vegetation restoration on carbonate‐derived laterite erodibility in karst mountain areas

AbstractVegetation restoration can exert a profound influence on surface soil characteristics, which likely plays a significant role in the evolution of soil erodibility. This study was conducted to quantify the effects of vegetation restoration on the soil erodibility of carbonate‐derived laterite in karst areas. Experiments were conducted at two representative sites for different years of vegetation restoration. Six karst mountain land use types [grassland (Y1), shrubland (Y2) and woodland (Y3) in young forest (YF); forest trails (O1), shrubland (O2) and woodland (O3) in old forest (OF)]. A structural equation model and random forest regression analysis were applied to distinguish the key environmental factors dominating soil erodibility. The results showed that the soil erodibility K factor exponentially decreased with increasing restoration age and tended to remain stable after 20 years (y = 0.0466 + 0.0316e−0.2889x, R2 = 0.98). The K factor in the OF decreased by an average of 2.1%–14.8% compared to those in the YF. A structural equation model indicated that vegetation restoration indirectly reduced soil erodibility by altering plant growth and coverage and its associated soil property changes. A random forest regression analysis of global datasets indicated that soil erodibility is mainly affected by topographical factors (i.e., latitude, elevation and slope), and their mean decrease accuracy values (IncMSE) exceeded 20%. Both forestland and shrubland were not considered as the most important factors for affecting global soil erodibility. Conversely, grassland, cropland and bare land were among the important factors. These results could guide the potential benefits of vegetation restoration strategies to meet soil conservation and ecosystem function goals in karst mountain ecosystems.

Developing Robust Flood Susceptibility Model with Small Numbers of Parameters in Highly Fertile Regions of Northwest Bangladesh for Sustainable Flood and Agriculture Management

The present study intends to improve the robustness of a flood susceptibility (FS) model with a small number of parameters in data-scarce areas, such as northwest Bangladesh, by employing machine learning-based sensitivity analysis and an analytical hierarchy process (AHP). In this study, the nine most relevant flood elements (such as distance from the river, rainfall, and drainage density) were chosen as flood conditioning variables for modeling. The FS model was produced using AHP technique. We used an empirical and binormal receiver operating characteristic (ROC) curves for validating the models. We performed Sensitivity analyses using a random forest (RF)-based mean Gini decline (MGD), mean decrease accuracy (MDA), and information gain ratio to find out the sensitive flood conditioning variables. After performing sensitivity analysis, the least sensitivity variables were eliminated. We re-ran the model with the rest of the parameters to enhance the model’s performance. Based on previous studies and the AHP weighting approach, the general soil type, rainfall, distance from river/canal (Dr), and land use/land cover (LULC) had higher factor weights of 0.22, 0.21, 0.19, and 0.15, respectively. The FS model without sensitivity and with sensitivity performed well in the present study. According to the RF-based sensitivity and information gain ratio, the most sensitive factors were rainfall, soil type, slope, and elevation, while curvature and drainage density were less sensitive parameters, which were excluded in re-running the FS model with just vital parameters. Using empirical and binormal ROC curves, the new FS model yields higher AUCs of 0.835 and 0.822, respectively. It is discovered that the predicted model’s robustness may be maintained or increased by removing less relevant factors. This study will aid decision-makers in developing flood management plans for the examined region.

Mean decrease accuracy for random forests: inconsistency, and a practical solution via the Sobol-MDA

Summary Variable importance measures are the main tools used to analyse the black-box mechanisms of random forests. Although the mean decrease accuracy is widely accepted as the most efficient variable importance measure for random forests, little is known about its statistical properties. In fact, the definition of mean decrease accuracy varies across the main random forest software. In this article, our objective is to rigorously analyse the behaviour of the main mean decrease accuracy implementations. Consequently, we mathematically formalize the various implemented mean decrease accuracy algorithms, and then establish their limits when the sample size increases. This asymptotic analysis reveals that these mean decrease accuracy versions differ as importance measures, since they converge towards different quantities. More importantly, we break down these limits into three components: the first two terms are related to Sobol indices, which are well-defined measures of a covariate contribution to the response variance, widely used in the sensitivity analysis field, as opposed to the third term, whose value increases with dependence within covariates. Thus, we theoretically demonstrate that the mean decrease accuracy does not target the right quantity to detect influential covariates in a dependent setting, a fact that has already been noticed experimentally. To address this issue, we define a new importance measure for random forests, the Sobol-mean decrease accuracy, which fixes the flaws of the original mean decrease accuracy, and consistently estimates the accuracy decrease of the forest retrained without a given covariate, but with an efficient computational cost. The Sobol-mean decrease accuracy empirically outperforms its competitors on both simulated and real data for variable selection.

Machine Learning Models Based on Random Forest Feature Selection and Bayesian Optimization for Predicting Daily Global Solar Radiation

Prediction of daily global solar radiation with simple and highly accurate models would be beneficial for solar energy conversion systems. In this paper, we proposed a hybrid machine learning methodology integrating two feature selection methods and a Bayesian optimization algorithm to predict H in the city of Fez, Morocco. First, we identified the most significant predictors using two Random Forest methods of feature importance: Mean Decrease in Impurity (MDI) and Mean Decrease in Accuracy (MDA). Then, based on the feature selection results, ten models were developed and compared: (1) five standalone machine learning (ML) models including Classification and Regression Trees (CART), Random Forests (RF), Bagged Trees Regression (BTR), Support Vector Regression (SVR), and Multi-Layer Perceptron (MLP); and (2) the same models tuned by the Bayesian optimization (BO) algorithm: CART-BO, RF-BO, BTR-BO, SVR-BO, and MLP-BO. Both MDI and MDA techniques revealed that extraterrestrial solar radiation and sunshine duration fraction were the most influential features. The BO approach improved the predictive accuracy of MLP, CART, SVR, and BTR models and prevented the CART model from overfitting. The best improvements were obtained using the MLP model, where RMSE and MAE were reduced by 17.6% and 17.2%, respectively. Among the studied models, the SVR-BO algorithm provided the best trade-off between prediction accuracy (RMSE=0.4473kWh/m²/day, MAE=0.3381kWh/m²/day, and R²=0.9465), stability (with a 0.0033kWh/m²/day increase in RMSE), and computational cost.

Abstract TP188: Predicting Ischemic Stroke And All-cause Mortality Risk In Patients With Heart Failure With Reduced Ejection Fraction And Sinus Rhythm: A Secondary Analysis Of The Warcef Trial

Introduction: There is a known risk of ischemic stroke (IS) and mortality among patients with heart failure with reduced ejection fraction (HFrEF) and sinus rhythm. We identify predictors of IS and all-cause death in patients with left ventricular ejection fraction (LVEF) < 35% and sinus rhythm by a machine-learning approach. Methods: We performed a post-hoc analysis of the WARCEF trial which randomized patients with LVEF < 35% and sinus rhythm to either aspirin or warfarin. We built two Random Forest for Survival, Longitudinal, and Multivariate (RF-SLAM) algorithms to model the outcomes of IS and all-cause death separately using all baseline variables with missingness < 10%. The training data set was comprised of 70% of observations and the testing set of the remaining 30%. To assess variable importance in outcome prediction, Mean Decrease Accuracy was computed for each variable over all out-of-bag cross-validated predictions. Results: In 2,298 participants (median age 61.5 years [IQR 54.3-69.6], 20% female) during a mean follow-up of 3.5 years (SD 1.8 years), 84 (3.7%) patients had an ischemic stroke and 547 (23.8%) died. Patient-specific characteristics contributing most to ischemic stroke prediction with variable Mean Decrease Accuracies ≥ 2.5 were prior transient ischemic attack, CHA 2 DS 2 -VASc score, age, prior stroke, glucose, and blood urea nitrogen (model accuracy 0.95; 95% C.I. 0.93-0.97). Age, hemoglobin, glucose, INR, platelet count, BUN, and systolic blood pressure were the most important patient-specific predictors of all-cause death with variable Mean Decrease Accuracies ≥ 2.5 (model accuracy 0.76; 95% C.I. 0.72-0.79). Conclusions: Clinical and serologic characteristics identified by the RF-SLAM algorithm predict IS and all-cause death risks in patients with HFrEF and sinus rhythm. Future studies are needed to assess the impact of these factors on treatment effects in heart failure trials evaluating outcomes which include stroke.

Classification of urban tree species using LiDAR data and WorldView-2 satellite imagery in a heterogeneous environment

Feature complexity and heterogeneity of urban areas pose a challenge for tree species classification. This study examined the effectiveness of the integrated Worldview-2 (WV-2) bands, vegetation indices and normalized Digital Surface Model (nDSM) dataset in mapping common urban tree species and other land use and land cover (LULC) types using Random Forest (RF) and Support Vector Machine (SVM) algorithms. The study also ranked the importance of nDSM, WV-2 bands and vegetation indices. The results indicate that the integrated dataset was effective as shown by high classification accuracies of 97% for the RF and 94% for SVM classifiers. The nDSM was the top-ranked variable with high Mean Decrease in Accuracy (MDA) and Mean Decrease in Gini (MDG) scores of 0.98 and 0.61, respectively. This research provides information to municipalities on the methods and data that can be used for the sustainable management of urban tree species.

Date of Admission during COVID-19 Pandemic Impacted Patient Outcomes in Addition to the Higher Efficacy of Tocilizumab Plus High-Dose Corticosteroid Therapy Compared to Tocilizumab Alone.

Background: The evidence for the efficacy of glucocorticoids combined with tocilizumab (TCZ) in COVID-19 comes from observational studies or subgroup analysis. Our aim was to compare outcomes between hospitalized COVID-19 patients who received high-dose corticosteroid pulse therapy and TCZ and those who received TCZ. Methods: A retrospective single-center study was performed on consecutive hospitalized patients with severe COVID-19 between 1 March and 23 April 2020. Patients treated with either TCZ (400–600 mg, one to two doses) and methylprednisolone pulses (MPD-TCZ group) or TCZ alone were analyzed for the occurrence of a combined endpoint of death and need for invasive mechanical ventilation during admission. The independence of both treatment groups was tested using machine learning classifiers, and relevant variables that were potentially different between the groups were measured through a mean decrease accuracy algorithm. Results: An earlier date of admission was significantly associated with worse outcomes regardless of treatment type. Twenty patients died (27.0%) in the TCZ group, and 33 (44.6%) died or required intubation (n = 74), whereas in the MPD-TCZ group, 15 (11.0%) patients died and 29 (21.3%) patients reached the combined endpoint (n = 136; p = 0.006 and p < 0.001, respectively). Machine learning methodology using a random forest classifier confirmed significant differences between the treatment groups. Conclusions: MPD and TCZ improved outcomes (death and invasive mechanical ventilation) among hospitalized COVID-19 patients, but confounding variables such as the date of admission during the COVID-19 pandemic should be considered in observational studies.

Identifying and characterizing patterns of functional decline in memory clinic patients

AbstractBackgroundAlzheimer’s disease and related disorders (ADRD) are characterized by progressive cognitive and functional decline. To improve prognosis, it is important to predict the expected rate of functional decline. In the current study, we sought to uncover patterns of changes in everyday functioning among memory clinic patients, and to explore factors associated with said groups.MethodMemory clinic patients (N=573, 64.7±7.9 years old, 39% female) underwent extensive dementia screening at baseline, and were followed over time (median 1.0 years, range 0.3–6.2 years, N observations=1632). Diagnoses included subjective cognitive decline, mild cognitive impairment, and dementia with varying etiologies. We measured everyday functioning with the Amsterdam Instrumental Activities of Daily Living Questionnaire (A‐IADL‐Q). Longitudinal latent class analysis (LLCA) was used to identify groups of patients based on their rate of change in IADL, including non‐linear trajectories. We subsequently used random forests to explore the associations between the LLCA groups and demographics, neuropsychological test performance, and magnetic resonance imaging rating scales for medial temporal and global atrophy.ResultWe identified six groups: two stable (high and average), three declining (slow, moderate, and fast), and one improving group (see Table 1, Figure 1). All groups differed in baseline A‐IADL‐Q scores, with the declining groups having lower scores. The fast declining group consisted mostly of dementia patients, whereas the high stable group consisted mostly of patients with subjective cognitive decline. The area under the curve of the random forest classification was .95, thus accurately assigning patients to the correct group based on baseline characteristics. The following variables contributed most to the classification of the groups: baseline A‐IADL‐Q score (mean decrease accuracy [MDA]=200.5), a visuospatial task (copy of Rey Complex Figure, MDA=36.6) and a processing speed task (Letter Digit Substitution Task, MDA=18.5). Atrophy rating scales contributed less.ConclusionDifferent trajectories of functional decline were distinguished, and associated with baseline levels of cognitive and daily functioning. Our findings suggest that neuropsychological assessment may provide additional information on future changes in daily functioning. These findings may contribute to a more accurate prognosis of future functional decline, and by extension, future care needs.

Classification and association rule mining of road collisions for analyzing the fatal severity, a case study

IntroductionNowadays, a significant part of goods and passengers are transported on suburban highways with mainly high-speed vehicles. Hence, these highways are very prone to collisions with different injuries. For this reason, road collisions have become one of the largest international health issues in the world recently. Due to the high fatality or severe physical/mental injury rates caused by car collisions and the complex interactions between the factors affecting them, analyzing these collision-prone areas, identifying the factors affecting their occurrences, and discovering knowledge in the form of key rules is crucial, as the purposes for this study. MethodsThree supervised algorithms including Artificial Neural Network (ANN), Support Vector Machine (SVM), and Random Forests (RF) were used to build up classification models for the fatality severity of 2355 fatal collision data records during 2007–2009 occurred in the roadways of 8 states in the USA with different driver-related, environmental, and road factors. Predicted risk maps were generated for each classifier and the importance of contributing factors was evaluated based on the mean decrease in accuracy and the mean decrease in Gini Index. Finally, association rule mining was performed by the Apriori algorithm to extract collision rules. ResultsRF outperformed the other methods in terms of the highest overall accuracy and kappa rates, which were 94% and 92% respectively. The risk maps revealed that collisions with the most fatalities were mainly concentrated in the northern states. The number of travel lanes, speed limit, roadway profile, and light conditions were the most influencing factors in the collisions. 68 association rules were mined by the Apriori algorithm. Then, important rules were specified to extract hidden information from the collision data. ConclusionsThe RF model was more accurate in collision analysis, along with comprehensive collision factors. Great importance should be given to road factors in route design, especially in critical locations with severe fatality risk.

Ensemble machine learning methods for spatio-temporal data analysis of plant and ratoon sugarcane

Recent technological enhancements in the field of information technology and statistical techniques allowed the sophisticated and reliable analysis based on machine learning methods. A number of machine learning data analytical tools may be exploited for the classification and regression problems. These tools and techniques can be effectively used for the highly data-intensive operations such as agricultural and meteorological applications, bioinformatics and stock market analysis based on the daily prices of the market. Machine learning ensemble methods such as Decision Tree (C5.0), Classification and Regression (CART), Gradient Boosting Machine (GBM) and Random Forest (RF) has been investigated in the proposed work. The proposed work demonstrates that temporal variations in the spectral data and computational efficiency of machine learning methods may be effectively used for the discrimination of types of sugarcane. The discrimination has been considered as a binary classification problem to segregate ratoon from plantation sugarcane. Variable importance selection based on Mean Decrease in Accuracy (MDA) and Mean Decrease in Gini (MDG) have been used to create the appropriate dataset for the classification. The performance of the binary classification model based on RF is the best in all the possible combination of input images. Feature selection based on MDA and MDG measures of RF is also important for the dimensionality reduction. It has been observed that RF model performed best with 97% accuracy, whereas the performance of GBM method is the lowest. Binary classification based on the remotely sensed data can be effectively handled using random forest method.

Evaluation of different DEMs for gully erosion susceptibility mapping using in-situ field measurement and validation

The spatial variability in any kind of geomorphic studies based on terrain attributes are the most important issues. This terrain attributes and their respective characteristics play a significant role in the formation and expansion of ephemeral gullies. Therefore, nowadays, the accuracy of terrain based geomorphic studies has been mostly dependent on the resolution and quality of the DEM (digital elevation model) data. As the rate of erosional power of water flow is dependent on the terrain characteristics, therefore the extraction of several terrain features from DEM data is necessary in the study of gully erosion. This case study investigates the scale-dependence of DEM-derived terrain factors in gully erosion susceptibility (GES). This work on Garhbeta-I C.D. Block has focused on the comparison among the predicted GES maps through five types of DEM i.e. Shuttle Radar Topography Mission (SRTM), Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), Cartosat-1, ALOS World 3D-30 m (AW3D30) and Advanced Land Observation satellite (ALOS) coupled with the machine learning modelling approach of artificial neural network (ANN), convolution neural network (CNN) and deep learning neural network (DLNN) algorithm. A total of sixteen conditioning factors were chosen for GES assessment based on the topographical, hydro-climatological conditions and multi-collinearity analysis. Here, importance variables are measured by mean decrease accuracy (MDA) method of random forest (RF) algorithm and the result is shown that elevation is the most important factor for gully occurrences. Validation result of receiver operating characteristics-area under curve (ROC-AUC) has been indicates that DLNN model in ALOS DEM (AUC = 0.958) gives the most optimal accuracy in GES assessment. The output maps can assist in identifying gully-prone risk areas, and several suitable with sustainable managements should be taken for conservation accordingly.

Sensitive Feature Evaluation for Soil Moisture Retrieval Based on Multi-Source Remote Sensing Data with Few In-Situ Measurements: A Case Study of the Continental U.S.

Soil moisture (SM) plays an important role for understanding Earth’s land and near-surface atmosphere interactions. Existing studies rarely considered using multi-source data and their sensitiveness to SM retrieval with few in-situ measurements. To solve this issue, we designed a SM retrieval method (Multi-MDA-RF) using random forest (RF) based on 29 features derived from passive microwave remote sensing data, optical remote sensing data, land surface models (LSMs), and other auxiliary data. To evaluate the importance of different features to SM retrieval, we first compared 10 filter or embedded type feature selection methods with sequential forward selection (SFS). Then, RF was employed to establish a nonlinear relationship between the in-situ SM measurements from sparse network stations and the optimal feature subset. The experiments were conducted in the continental U.S. (CONUS) using in-situ measurements during August 2015, with only 5225 training samples covering the selected feature subset. The experimental results show that mean decrease accuracy (MDA) is better than other feature selection methods, and Multi-MDA-RF outperforms the back-propagation neural network (BPNN) and generalized regression neural network (GRNN), with the R and unbiased root-mean-square error (ubRMSE) values being 0.93 and 0.032 cm3/cm3, respectively. In comparison with other SM products, Multi-MDA-RF is more accurate and can well capture the SM spatial dynamics.

Factors contributing to successful employment outcomes for individuals who are hard-of-hearing

BACKGROUND: The State-Federal Vocational Rehabilitation (VR) Program provides rehabilitation services to people with disabilities with the intention of assisting them in securing competitive employment. The VR services offer substantial resources to help individuals who are hard-of-hearing to enhance their quality of life and employment opportunities. OBJECTIVE: The current study investigated the impact of demographic variables and the use of VR services on employment outcomes among hard-of-hearing consumers. Specific VR services that lead to successful employment among hard-of-hearing consumers were thoroughly examined. METHODS: Binary logistic regression, Chi-square, and Chi-square Automatic Interaction Detector analyses were used to analyze the data extracted from the 2014 fiscal year US. Department of Education Rehabilitation Service Administration Case Service Report (RSA-911). RESULTS: Logistic regression reveals that VR services such as diagnosis and treatment of impairments (p-value 0.000), counseling, and guidance (p-value 0.000), and rehabilitation technology (p-value 0.000) were influential factors in determining the successful employment outcome among the consumers. The relative importance of the factors based on the mean decrease in accuracy in CHAID identifies rehabilitation technology (0.264), diagnosis and treatment of impairments (0.090), job placement assistance (0.016), transportation (0.016), and secondary disability (0.010) to be among the most contributing factors. CONCLUSION: Overall, rehabilitation technology services were especially beneficial, particularly for minority consumers, in achieving a successful employment outcome.

Improving Performance Loan Fraud Model Prediction Using Mean Decrease Accuracy and Mean Decrease Gini

The purpose of this research is to develop a fraud detection model on loan transactions at failed banks in the context of deposit and deposit guarantees mandated to the Indonesia Deposit Insurance Corporation (IDIC). The data used in this study is the data of a bank in the Jakarta area that had liquidated at the end of 2015. Meanwhile, data on loan transaction ranges ranged from 2010 to 2015. This research also focuses on improving the performance of detection models by using feature selection. With the feature selection, it expected that the impact of the reduced performance of the model exposed to high variance and high bias due to the many features used can handled better.

Sentinel-1 and 2 Time-Series for Vegetation Mapping Using Random Forest Classification: A Case Study of Northern Croatia

Land-cover (LC) mapping in a morphologically heterogeneous landscape area is a challenging task since various LC classes (e.g., crop types in agricultural areas) are spectrally similar. Most research is still mostly relying on optical satellite imagery for these tasks, whereas synthetic aperture radar (SAR) imagery is often neglected. Therefore, this research assessed the classification accuracy using the recent Sentinel-1 (S1) SAR and Sentinel-2 (S2) time-series data for LC mapping, especially vegetation classes. Additionally, ancillary data, such as texture features, spectral indices from S1 and S2, respectively, as well as digital elevation model (DEM), were used in different classification scenarios. Random Forest (RF) was used for classification tasks using a proposed hybrid reference dataset derived from European Land Use and Coverage Area Frame Survey (LUCAS), CORINE, and Land Parcel Identification Systems (LPIS) LC database. Based on the RF variable selection using Mean Decrease Accuracy (MDA), the combination of S1 and S2 data yielded the highest overall accuracy (OA) of 91.78%, with a total disagreement of 8.22%. The most pertinent features for vegetation mapping were GLCM Mean and Variance for S1, NDVI, along with Red and SWIR band for S2, whereas the digital elevation model produced major classification enhancement as an input feature. The results of this study demonstrated that the aforementioned approach (i.e., RF using a hybrid reference dataset) is well-suited for vegetation mapping using Sentinel imagery, which can be applied for large-scale LC classifications.

Identification of Ten Core Hub Genes as Potential Biomarkers and Treatment Target for Hepatoblastoma.

BackgroundThis study aimed to systematically investigate gene signatures for hepatoblastoma (HB) and identify potential biomarkers for its diagnosis and treatment.Materials and MethodsGSE131329 and GSE81928 were obtained from the Gene Expression Omnibus (GEO) database. Differentially expressed genes (DEGs) between hepatoblastoma and normal samples were identified using the Limma package in R. Then, the similarity of network traits between two sets of genes was analyzed by weighted gene correlation network analysis (WGCNA). Cytoscape was used to visualize and select hub genes. PPI network of hub genes was construed by Cytoscape. GO enrichment and KEGG pathway analyses of hub genes were carried out using ClueGO. The random forest classifier was constructed based on the hub genes using the GSE131329 dataset as the training set, and its reliability was validated using the GSE81928 dataset. The resulting core hub genes were combined with the InnateDB database to identify the innate core genes.ResultsA total of 4244 DEGs in HB were identified. WGCNA identified four modules that were significantly correlated with the disease status. A total of 114 hub genes were obtained within the top 20 genes of each node rank. 6982 relation pairs and 3700 nodes were contained in the PPI network of 114 hub genes. GO enrichment and KEGG pathway analyses of hub genes were focused on MAPK, cell cycle, p53, and other crucial pathways involved in HB. A random forest classifier was constructed using the 114 hub genes as feature genes, resulting in a 95.5% true positive rate when classifying HB and normal samples. A total of 35 core hub genes were obtained through the mean decrease in accuracy and mean decrease Gini of the random forest model. The classification efficiency of the random forest model was 81.4%. Finally, CDK1, TOP2A, ADRA1A, FANCI, XRCC1, TPX2, CCNB2, CDK4, GLYATL1, and CFHR3 were identified by cross-comparison with the InnateDB database.ConclusionOur study established a random forest classifier that identified 10 core genes in HB. These findings may be beneficial for the diagnosis, prediction, and targeted therapy of HB.

High-Dimensional Satellite Image Compositing and Statistics for Enhanced Irrigated Crop Mapping

Accurate irrigated area maps remain difficult to generate, as smallholder irrigation schemes often escape detection. Efforts to map smallholder irrigation have often relied on complex classification models fitted to temporal image stacks. The use of high-dimensional geometric median composites (geomedians) and high-dimensional statistics of time-series may simplify classification models and enhance accuracy. High-dimensional statistics for temporal variation, such as the spectral median absolute deviation, indicate spectral variability within a period contributing to a geomedian. The Ord River Irrigation Area was used to validate Digital Earth Australia’s annual geomedian and temporal variation products. Geomedian composites and the spectral median absolute deviation were then calculated on Sentinel-2 images for three smallholder irrigation schemes in Matabeleland, Zimbabwe, none of which were classified as areas equipped for irrigation in AQUASTAT’s Global Map of Irrigated Areas. Supervised random forest classification was applied to all sites. For the three Matabeleland sites, the average Kappa coefficient was 0.87 and overall accuracy was 95.9% on validation data. This compared with 0.12 and 77.2%, respectively, for the Food and Agriculture Organisation’s Water Productivity through Open access of Remotely sensed derived data (WaPOR) land use classification map. The spectral median absolute deviation was ranked among the most important variables across all models based on mean decrease in accuracy. Change detection capacity also means the spectral median absolute deviation has some advantages for cropland mapping over indices such as the Normalized Difference Vegetation Index. The method demonstrated shows potential to be deployed across countries and regions where smallholder irrigation schemes account for large proportions of irrigated area.

Improving Spatial Coverage of Satellite Aerosol Classification Using a Random Forest Model

The spatial coverage of satellite aerosol classification was improved using a random forest (RF) model trained with observational data including target (aerosol type) and input (satellite measurement) variables. The AErosol RObotic NETwork (AERONET) aerosol-type dataset was used for the target variables. Satellite input variables with many missing data or low mean-decrease accuracy were excluded from the final input variable set, and good performance in aerosol-type classification was achieved. The performance of the RF-based model was evaluated on the basis of the wavelength dependence of single-scattering albedo (SSA) and fine-mode-fraction values from AERONET. Typical SSA wavelength dependence for individual aerosol types was consistent with that obtained for aerosol types by the RF-based model. The spatial coverage of the RF-based model was also compared with that of previously developed models in a global-scale case study. The study demonstrates that the RF-based model allows satellite aerosol classification with improved spatial coverage, with a performance similar to that of previously developed models.

Orbita hyperspectral satellite image for land cover classification using random forest classifier

The Orbita hyperspectral satellite (OHS) is the first commercial hyperspectral satellite in China that completed launching and networking. It can collect world-class hyperspectral data and obtain aerial hyperspectral imagery with 32 bands covering the spectrum range from 400 to 1000 nm at a 10-m resolution, which are of great significance for the quantitative analyses of remote sensing and refined recognitions of land covers. We explore the potentiality of the OHS image in land cover classification (LCC). Taking the Pearl River Delta region as the study area, we selected five feature indices from OHS data, i.e., original bands (OBs), vegetation indices (VIs), water indices (WIs), red edge indices (REIs), and gray-level co-occurrence matrix (GLCM) textures for the LCC. Then, data combination schemes were intended to analyze and compare the performance of different feature indices on the accuracies of the LCC. Last, feature optimization was performed on all input variables to determine the optimal variables combination to increase the accuracy and efficiency of the LCC. The random forest classifier was adopted in the above schemes, and the method of mean decrease in accuracy was used to determine the importance of the variables. The results show that (1) refined accuracy of LCC was obtained using only OBs; in addition, REIs can further improve the classification accuracy significantly. (2) The optimal variables combination achieves the highest classification accuracy (OA = 93.21 % and kappa coefficient = 0.91), and the user’s and producer’s accuracies exceeded 90% for most land cover categories. (3) Variable importance analyses show that the importance of both red-edge bands and REIs were greater than those of near-infrared bands and VIs for the LCC. The importance ranking of different indices from high to low was REIs > OBs > GLCM > WIs > VIs. This research demonstrates the potentialities and values of the OHS image for the application of LCC.