An evaluation of Guided Regularized Random Forest for classification and regression tasks in remote sensing

Izquierdo‐Verdiguier, Emma; Zurita-Milla, Raúl

doi:10.1016/j.jag.2020.102051

Cited by 81 publications

(43 citation statements)

References 47 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In general, the overall accuracies achieved were above 90% for all scenarios (combined and selected variables using GRRF). However, the higher accuracies of all combined scenarios (C1-C8) have been treated with caution because of the effects of multicollinearity that can lead to overfitting [58]. We thus placed more emphasis on the performance of the GRRF selected scenarios (C1selected-C8selected) which have the capabilities of reducing multidimensionality and the expected multicollinearity by maintaining the most relevant variables for the analysis.…”

Section: Discussionmentioning

confidence: 99%

“…To counter for the limitations of RRF, GRRF assigns a penalty coefficient to each feature by changing the importance coefficient of gamma value (γ ∈ (0, 1)), which controls the weight of normalized importance. This ensures that the most relevant variables are retained while still ensuring accurate classification [58]. Thus, in this study γ=0.7 was used in the "CoefReg" function in the "caret" package in R software to select the most important variables [59].…”

Section: Predictor Variables Selection and Classificationmentioning

confidence: 99%

See 1 more Smart Citation

Sophisticated Cropping Patterns Mapping in Kenya Using Vegetation Indices and Phenology from Sentinel-2 and Sentinel-1 Radar Backscatter Data

Aduvukha¹,

Abdel‐Rahman²,

Sichangi³

et al. 2020

Preprint

View full text Add to dashboard Cite

The proportion of area under various crops at a given point in time, known as a cropping pattern, plays an essential role in determining the level of agricultural production. In this study, cropping patterns of three sub-counties in Murang’a County, a typical African smallholder farming area in Kenya, were mapped. Specifically, we compared the performance of eight classification scenarios for mapping cropping patterns; namely using (i) only Sentinel-2 reflectance bands (S2), (ii) S2 and S2 derived vegetation indices (VIs); (iii) S2 and S2 vegetation phenology (VP); (iv) S2 and Sentinel-1 radar backscatter data (S1); (v) S2, VIs, and S1; (vi) S2, VP, and S1; (vii) S2, VIs and VP, and (viii) S2, VIs, VP and S1. Reference data of the dominant cropping patterns and non-croplands were collected. The guided regularized random forest (GRRF) algorithm was used to select the optimum variables and to perform the respective classification for each scenario. The most accurate result of the overall accuracy of 93.16% was attained from the scenario (viii) S2, VIs, VP, and S1. The McNemar’s test of significance did not show significant differences (p≤0.05) among the tested scenarios. Our study demonstrated the strength of GRRF and the synergetic advantage of S2 and S1 derivatives to map cropping patterns in a heterogeneous landscape where high resolution imagery are inaccessible. Our cropping pattern mapping approach can be used in other sites of relatively similar agro-ecological conditions.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Predictor Variables Selection and Classificationmentioning

confidence: 99%

Sophisticated Cropping Patterns Mapping in Kenya Using Vegetation Indices and Phenology from Sentinel-2 and Sentinel-1 Radar Backscatter Data

Aduvukha¹,

Abdel‐Rahman²,

Sichangi³

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…It was necessary to reduce the number of features used to train and apply the classifiers to ensure timely execution of those classifiers in GEE. To conduct feature reduction, we applied a Guided Regularized Random Forest (GRRF [39]) that is a feature selection algorithm based in Random Forest [40]. Then, we applied a Support Vector Machine [41] classifier with a Gaussian kernel to the top ten most accurate subsets of features returned by GRRF.…”

Section: Data Pre-processingmentioning

confidence: 99%

“…This approach selected the most informative feature for our classification task among similar features, e.g., NDVI and EVI. GRRF does not require a prior parameter setup and ensures the selection of non-redundant and informative features improving the classification accuracy when applying a Random Forest algorithm afterwards, although a small negative impact on accuracy has been reported for regression tasks [39]. On the other hand, SVM is a robust and widely applied machine learning algorithm that generalizes well when the number of input features is large, and the training data are limited [41,42].…”

Section: Data Pre-processingmentioning

confidence: 99%

Cloud Computation Using High-Resolution Images for Improving the SDG Indicator on Open Spaces

Aguilar

Kuffer

2020

Remote Sensing

View full text Add to dashboard Cite

Open spaces are essential for promoting quality of life in cities. However, accelerated urban growth, in particular in cities of the global South, is reducing the often already limited amount of open spaces with access to citizens. The importance of open spaces is promoted by SDG indicator 11.7.1; however, data on this indicator are not readily available, neither globally nor at the metropolitan scale in support of local planning, health and environmental policies. Existing global datasets on built-up areas omit many open spaces due to the coarse spatial resolution of input imagery. Our study presents a novel cloud computation-based method to map open spaces by accessing the multi-temporal high-resolution imagery repository of Planet. We illustrate the benefits of our proposed method for mapping the dynamics and spatial patterns of open spaces for the city of Kampala, Uganda, achieving a classification accuracy of up to 88% for classes used by the Global Human Settlement Layer (GHSL). Results show that open spaces in the Kampala metropolitan area are continuously decreasing, resulting in a loss of open space per capita of approximately 125 m2 within eight years.

show abstract

“…Precision, sensitivity, specificity, G-Mean= sqrt (Sensitivity×Specificity), F1-Score, area under ROC curve AUC and other parameters together were used as predictive evaluation indicators [22] . In Table 3 and Figure 4, 7 classification models are selected for comparison, covering probability model, tree model, linear model, ensemble model and neural network model.It comprehensively reflects the performance of the research objects in different classification models and the ensemble model has the best and most stable effect, in this paper.…”

Section: Comprehensive Evaluation Indexmentioning

confidence: 99%

Artificial Intelligence Applications in allergic rhinitis diagnosis: Focus on Ensemble Learning

Dai¹,

Meng

Tan

et al. 2020

Preprint

View full text Add to dashboard Cite

Background and purpose: Artificial intelligence is an important product of the rapid development of computer technology today. This study intends to propose an intelligent diagnosis and detection method for AR based on ensemble learning. Method: This study collectedAR cases and other 7 types of diseases with similar symptoms:Rhinosinusitis, Chronic rhinitis, upper respiratory tract infection etc.) and collected clinical data such as medical history, clinical symptoms, allergen detection and imaging. Multiple models are used to train the classifier for the same batch of data, and the final ensemble classifier is obtained by using the ensemble learning algorithm. 5 common machine learning classification algorithms were selected for comparative experiments, including Naive Bayes (NB), Support Vector Machine (SVM), Logistic Regression (LR), Multilayer Perceptron (MLP), Deep Forest (GCForest), eXtreme Gradient boosting (XGBoost). In order to evaluate the prediction results of AR samples, parameters such as Precision, Sensitivity, Specificity, G-Mean, F1-Score, and AUC under the ROC curve are jointly used as prediction evaluation indicators. Results: 7 classification models are used for comparison, covering probability model, tree model, linear model, ensemble model and neural network models, and the comprehensive classification evaluation index is lower than the ensemble classification algorithms ARF-OOBEE and GCForest. Compared with other algorithms, the accuracy of G-Mean and AUC parameters is improved nearly 2%, and it has good comprehensive classification characteristics for massive large data and unbalanced samples. Conclusion: The ensemble learning ARF-OOBEE model has good generalization performance and comprehensive classification ability to be used for diagnosis of AR.

show abstract

An evaluation of Guided Regularized Random Forest for classification and regression tasks in remote sensing

Cited by 81 publications

References 47 publications

Sophisticated Cropping Patterns Mapping in Kenya Using Vegetation Indices and Phenology from Sentinel-2 and Sentinel-1 Radar Backscatter Data

Sophisticated Cropping Patterns Mapping in Kenya Using Vegetation Indices and Phenology from Sentinel-2 and Sentinel-1 Radar Backscatter Data

Cloud Computation Using High-Resolution Images for Improving the SDG Indicator on Open Spaces

Artificial Intelligence Applications in allergic rhinitis diagnosis: Focus on Ensemble Learning

Contact Info

Product

Resources

About