Investigating the Utility of Oblique Tree-Based Ensembles for the Classification of Hyperspectral Data

Accelerated and Interpretable Oblique Random Survival Forests

The oblique random survival forest (RSF) is an ensemble supervised learning method for right-censored outcomes. Trees in the oblique RSF are grown using linear combinations of predictors, whereas in the standard RSF, a single predictor is used. Oblique RSF ensembles have high prediction accuracy, but assessing many linear combinations of predictors induces high computational overhead. In addition, few methods have been developed for estimation of variable importance (VI) with oblique RSFs. We introduce a method to increase computational efficiency of the oblique RSF and a method to estimate VI with the oblique RSF. Our computational approach uses Newton-Raphson scoring in each non-leaf node, We estimate VI by negating each coefficient used for a given predictor in linear combinations, and then computing the reduction in out-of-bag accuracy. In benchmarking experiments, we find our implementation of the oblique RSF is hundreds of times faster, with equivalent prediction accuracy, compared to existing software for oblique RSFs. We find in simulation studies that "negation VI" discriminates between relevant and irrelevant numeric predictors more accurately than permutation VI, Shapley VI, and a technique to measure VI using analysis of variance. All oblique RSF methods in the current study are available in the aorsf R package, and additional supplemental materials are available online.

Predicting Water Stress in Wild Blueberry Fields Using Airborne Visible and Near Infrared Imaging Spectroscopy

Water management and irrigation practices are persistent challenges for many agricultural systems, exacerbated by changing seasonal and weather patterns. The wild blueberry industry is at heightened susceptibility due to its unique growing conditions and uncultivated nature. Stress detection in agricultural fields can prompt management responses to mitigate detrimental conditions, including drought and disease. We assessed airborne spectral data accompanied by ground sampled water potential over three developmental stages of wild blueberries collected throughout the 2019 summer on two adjacent fields, one irrigated and one non-irrigated. Ground sampled leaves were collected in tandem to the hyperspectral image collection with an unoccupied aerial vehicle (UAV) and then measured for leaf water potential. Using methods in machine learning and statistical analysis, we developed models to determine irrigation status and water potential. Seven models were assessed in this study, with four used to process six hyperspectral cube images for analysis. These images were classified as irrigated or non-irrigated and estimated for water potential levels, resulting in an R2 of 0.62 and verified with a validation dataset. Further investigation relating imaging spectroscopy and water potential will be beneficial in understanding the dynamics between the two for future studies.

Analysis of near infrared spectra for age-grading of wild populations of Anopheles gambiae

BACKGROUND

Understanding the age-structure of mosquito populations, especially malaria vectors such as Anopheles gambiae, is important for assessing the risk of infectious mosquitoes, and how vector control interventions may impact this risk. The use of near-infrared spectroscopy (NIRS) for age-grading has been demonstrated previously on laboratory and semi-field mosquitoes, but to date has not been utilized on wild-caught mosquitoes whose age is externally validated via parity status or parasite infection stage. In this study, we developed regression and classification models using NIRS on datasets of wild An. gambiae (s.l.) reared from larvae collected from the field in Burkina Faso, and two laboratory strains. We compared the accuracy of these models for predicting the ages of wild-caught mosquitoes that had been scored for their parity status as well as for positivity for Plasmodium sporozoites.

RESULTS

Regression models utilizing variable selection increased predictive accuracy over the more common full-spectrum partial least squares (PLS) approach for cross-validation of the datasets, validation, and independent test sets. Models produced from datasets that included the greatest range of mosquito samples (i.e. different sampling locations and times) had the highest predictive accuracy on independent testing sets, though overall accuracy on these samples was low. For classification, we found that intramodel accuracy ranged between 73.5-97.0% for grouping of mosquitoes into "early" and "late" age classes, with the highest prediction accuracy found in laboratory colonized mosquitoes. However, this accuracy was decreased on test sets, with the highest classification of an independent set of wild-caught larvae reared to set ages being 69.6%.

CONCLUSIONS

Variation in NIRS data, likely from dietary, genetic, and other factors limits the accuracy of this technique with wild-caught mosquitoes. Alternative algorithms may help improve prediction accuracy, but care should be taken to either maximize variety in models or minimize confounders.

Hyperspectral Image Classification with Spatial Filtering and \(l_{(2,1)}\) Norm

Recently, the sparse representation based classification methods have received particular attention in the classification of hyperspectral imagery. However, current sparse representation based classification models have not considered all the test pixels simultaneously. In this paper, we propose a hyperspectral classification method with spatial filtering and ℓ2,1 norm (SFL) that can deal with all the test pixels simultaneously. The ℓ2,1 norm regularization is used to extract relevant training samples among the whole training data set with joint sparsity. In addition, the ℓ2,1 norm loss function is adopted to make it robust for samples that deviate significantly from the rest of the samples. Moreover, to take the spatial information into consideration, a spatial filtering step is implemented where all the training and testing samples are spatially averaged with its nearest neighbors. Furthermore, the non-negative constraint is added to the sparse representation matrix motivated by hyperspectral unmixing. Finally, the alternating direction method of multipliers is used to solve SFL. Experiments on real hyperspectral images demonstrate that the proposed SFL method can obtain better classification performance than some other popular classifiers.