Optimal Random Forest Research Articles

Application of random forest, generalised linear model and their hybrid methods with geostatistical techniques to count data: Predicting sponge species richness

Spatial distribution of sponge species richness (SSR) and its relationship with environment are important for marine ecosystem management, but they are either unavailable or unknown. Hence we applied random forest (RF), generalised linear model (GLM) and their hybrid methods with geostatistical techniques to SSR data by addressing relevant issues with variable selection and model selection. It was found that: 1) of five variable selection methods, one is suitable for selecting optimal RF predictive models; 2) traditional model selection methods are unsuitable for identifying GLM predictive models and joint application of RF and AIC can select accuracy-improved models; 3) highly correlated predictors may improve RF predictive accuracy; 4) hybrid methods for RF can accurately predict count data; and 5) effects of model averaging are method-dependent. This study depicted the non-linear relationships of SSR and predictors, generated spatial distribution of SSR with high accuracy and revealed the association of high SSR with hard seabed features.

Selecting Optimal Random Forest Predictive Models: A Case Study on Predicting the Spatial Distribution of Seabed Hardness.

Spatially continuous predictions of seabed hardness are important baseline environmental information for sustainable management of Australia’s marine jurisdiction. Seabed hardness is often inferred from multibeam backscatter data with unknown accuracy and can be inferred from underwater video footage at limited locations. In this study, we classified the seabed into four classes based on two new seabed hardness classification schemes (i.e., hard90 and hard70). We developed optimal predictive models to predict seabed hardness using random forest (RF) based on the point data of hardness classes and spatially continuous multibeam data. Five feature selection (FS) methods that are variable importance (VI), averaged variable importance (AVI), knowledge informed AVI (KIAVI), Boruta and regularized RF (RRF) were tested based on predictive accuracy. Effects of highly correlated, important and unimportant predictors on the accuracy of RF predictive models were examined. Finally, spatial predictions generated using the most accurate models were visually examined and analysed. This study confirmed that: 1) hard90 and hard70 are effective seabed hardness classification schemes; 2) seabed hardness of four classes can be predicted with a high degree of accuracy; 3) the typical approach used to pre-select predictive variables by excluding highly correlated variables needs to be re-examined; 4) the identification of the important and unimportant predictors provides useful guidelines for further improving predictive models; 5) FS methods select the most accurate predictive model(s) instead of the most parsimonious ones, and AVI and Boruta are recommended for future studies; and 6) RF is an effective modelling method with high predictive accuracy for multi-level categorical data and can be applied to ‘small p and large n’ problems in environmental sciences. Additionally, automated computational programs for AVI need to be developed to increase its computational efficiency and caution should be taken when applying filter FS methods in selecting predictive models.

Single and Multi-Date Landsat Classifications of Basalt to Support Soil Survey Efforts

Basalt outcrops are significant features in the Western United States and consistently present challenges to Natural Resources Conservation Service (NRCS) soil mapping efforts. Current soil survey methods to estimate basalt outcrops involve field transects and are impractical for mapping regionally extensive areas. The purpose of this research was to investigate remote sensing methods to effectively determine the presence of basalt rock outcrops. Five Landsat 5 TM scenes (path 39, row 29) over the year 2007 growing season were processed and analyzed to detect and quantify basalt outcrops across the Clark Area Soil Survey, ID, USA (4,570 km2). The Robust Classification Method (RCM) using the Spectral Angle Mapper (SAM) method and Random Forest (RF) classifications was applied to individual scenes and to a multitemporal stack of the five images. The highest performing RCM basalt classification was obtained using the 18 July scene, which yielded an overall accuracy of 60.45%. The RF classifications applied to the same datasets yielded slightly better overall classification rates when using the multitemporal stack (72.35%) than when using the 18 July scene (71.13%) and the same rate of successfully predicting basalt (61.76%) using out-of-bag sampling. For optimal RCM and RF classifications, uncertainty tended to be lowest in irrigated areas; however, the RCM uncertainty map included more extensive areas of low uncertainty that also encompassed forested hillslopes and riparian areas. RCM uncertainty was sensitive to the influence of bright soil reflectance, while RF uncertainty was sensitive to the influence of shadows. Quantification of basalt requires continued investigation to reduce the influence of vegetation, lichen and loess on basalt detection. With further development, remote sensing tools have the potential to support soil survey mapping of lava fields covering expansive areas in the Western United States and other regions of the world with similar soilscapes.

Rice Seed Cultivar Identification Using Near-Infrared Hyperspectral Imaging and Multivariate Data Analysis

A near-infrared (NIR) hyperspectral imaging system was developed in this study. NIR hyperspectral imaging combined with multivariate data analysis was applied to identify rice seed cultivars. Spectral data was exacted from hyperspectral images. Along with Partial Least Squares Discriminant Analysis (PLS-DA), Soft Independent Modeling of Class Analogy (SIMCA), K-Nearest Neighbor Algorithm (KNN) and Support Vector Machine (SVM), a novel machine learning algorithm called Random Forest (RF) was applied in this study. Spectra from 1,039 nm to 1,612 nm were used as full spectra to build classification models. PLS-DA and KNN models obtained over 80% classification accuracy, and SIMCA, SVM and RF models obtained 100% classification accuracy in both the calibration and prediction set. Twelve optimal wavelengths were selected by weighted regression coefficients of the PLS-DA model. Based on optimal wavelengths, PLS-DA, KNN, SVM and RF models were built. All optimal wavelengths-based models (except PLS-DA) produced classification rates over 80%. The performances of full spectra-based models were better than optimal wavelengths-based models. The overall results indicated that hyperspectral imaging could be used for rice seed cultivar identification, and RF is an effective classification technique.

Classification of bioaccumulative and non-bioaccumulative chemicals using statistical learning approaches

The present work aimed at developing in silico models allowing for a reliable prediction of bioaccumulative compounds and non-bioaccumulative compounds based on the definition of Bioconcentration Factor (BCF) using a diverse data set of 238 organic molecules. The partial least squares analysis (PLS), C4.5, support vector machine (SVM), and random forest (RF) algorithms were applied, and their performance classifying these compounds in terms of their quantitative structure-activity relationships (QSAR) was evaluated and verified with 5-fold cross-validation and an independent evaluation data set. The obtained results show that the overall prediction accuracies (Q) of the optimal PLS, C4.5, SVM and RF models are 84.5-87.7% for the internal cross-validation, with prediction accuracy (CO) of 86.3-91.1% in the external test sets, and C4.5 is slightly better than the three other methods which presents a Q of 87.7%, and a CO of 91.1% for the test sets. All these results prove the reliabilities of the in silico models, which should be valuable for the environmental risk assessment of the substances.