Detection of Ionospheric Scintillation Based on XGBoost Model Improved by SMOTE-ENN Technique

Abstract

Ionospheric scintillation frequently occurs in equatorial, auroral and polar regions, posing a threat to the performance of the global navigation satellite system (GNSS). Thus, the detection of ionospheric scintillation is of great significance in regard to improving GNSS performance, especially when severe ionospheric scintillation occurs. Normal algorithms exhibit insensitivity in strong scintillation detection in that the natural phenomenon of strong scintillation appears only occasionally, and such samples account for a small proportion of the data in datasets relative to those for weak/moderate scintillation events. Aiming at improving the detection accuracy, we proposed a strategy combining an improved eXtreme Gradient Boosting (XGBoost) algorithm by using the synthetic minority, oversampling technique and edited nearest neighbor (SMOTE-ENN) resampling technique for detecting events imbalanced with respect to weak, medium and strong ionospheric scintillation. It outperformed the decision tree and random forest by 12% when using imbalanced training and validation data, for tree depths ranging from 1 to 30. For different degrees of imbalance in the training datasets, the testing accuracy of the improved XGBoost was about 4% to 5% higher than that of the decision tree and random forest. Meanwhile, the testing results for the improved method showed significant increases in evaluation indicators, while the recall value for strong scintillation events was relatively stable, above 90%, and the corresponding F1 scores were over 92%. When testing on datasets with different degrees of imbalance, there was a distinct increase of about 10% to 20% in the recall value and 6% to 11% in the F1 score for strong scintillation events, with the testing accuracy ranging from 90.42% to 96.04%.

Detecting High-Risk Factors and Early Diagnosis of Diabetes Using Machine Learning Methods

“…Although the data has no missing values, the dataset was extremely imbalanced, as shown in Figure 2 . In an imbalanced data scenario, the data of a certain type are fewer in number than the other types of data in a dataset [ 25 ]. Most of the time, the minority class type is of interest for investigation.…”

“…In this study, the imbalanced dataset problem was handled using SMOTE-ENN. SMOTE-ENN [ 28 ] is a powerful method that merges the advantages of both SMOTE and ENN, with SMOTE oversampling the minority class and ENN undersampling the majority class samples [ 25 ]. Moreover, ENN drops any samples whose class types are different from the class of at least two of its three nearest neighbors; hence, any sample that is inaccurately classified by its three nearest neighbors is dropped from the training dataset [ 29 ].…”

“…[34] in 2016. is is an enhanced algorithm based on gradient boosting DT that can significantly build boosted trees and execute them in parallel [35]. In the iteration process, gradient boosting seeks to enhance the robustness by dropping the loss function of the algorithm as well as the gradient direction [25]. XGB trains multiple classifiers slowly and sequentially.…”

The determinants of investment fraud: A machine learning and artificial intelligence approach

“…The SMOTe+ENN technique works by oversampling the minority class and then editing the resulting dataset so that any samples too close to the boundary between classes are removed (Xu et al, 2020 ). The resultant effect is a dataset more representative of the true class distribution and less likely to overgeneralize (Lin et al, 2021 ).…”

Research and application of XGBoost in imbalanced data

“…The extreme gradient boosting algorithm XGBoost 8 is an ensemble learning algorithm with the advantages of high flexibility, strong predictability, strong generalization ability, high scalability, high model training efficiency, and great robustness. The current research work on XGBoost mainly focuses on direct application, [9][10][11][12][13][14] integration with other algorithms, [15][16][17][18] and parameter optimization. [19][20][21] In terms of imbalanced data research, Jia 22 combined the improved SMOTE algorithm of clustering with XGBoost, and applied ensemble learning to realize the abnormal detection of bolt process.…”