Support Vector Classification Algorithm Research Articles

Support Vector Machine based Data Hacking Prediction using PMU Data

As global reliance on power systems grows due to increasing energy demands and modern consumption patterns, maintaining the stability and reliability of the power grid has become crucial. Power systems are complex and nonlinear, and their operations are continuously evolving, making it difficult and expensive to ensure stability. Traditionally, power systems are designed to handle a single outage at a time. However, recent years have seen several significant blackouts, each originating from a single failure, which have been extensively reported. These reports are vital for mitigating operational risks by strengthening systems against identified high-risk scenarios. While extensive research has been conducted on these blackouts, cyber- attacks introduce a new dimension of risk. The advent of Phasor Measurement Units (PMUs) has enabled centralized monitoring of power system data, allowing for more effective fault and cyber-attack detection.This paper proposes a machine learning-based approach to detecting cyber-attacks using PMU data. Given the complexity and volume of power system data, traditional mathematical and statistical methods are challenging to implement. Instead, a Support Vector Classification (SVC) algorithm is used for binary classification, distinguishing between 'attack' and 'normal' states. The algorithm is trained on PMU data and evaluated using metrics such as the AUC-ROC curve and confusion matrix, achieving an 82% AUC- ROC score, demonstrating its effectiveness in identifying cyber- attacks.

The credibility assessment of Twitter/X users based organization objectives by heterogeneous resources in big data life cycle

Social network data, such as Twitter/X, is of Big Social Data type. Big social data describes people's social behaviors and interactions. They have high business value for decision-making in organizations. However, because of the anonymous nature of social network users, their credibility is ambiguous. Credibility expresses the accuracy and value of big social data. Despite extensive research on the credibility of big social data, most methods have not paid sufficient attention to the important dimensions of their assessment, including user expertise based on topic, selecting social network features, and labeling them. Furthermore, these methods cannot manage the time, high volume, and speed of big social data. To address these issues, this paper presents a novel model for assessing the credibility of Twitter/X users by integrating Twitter/X with Google Scholar. The model automatically defines users' credibility labels using Google Scholar. Machine learning feature selection methods also select features that affect the credibility of Twitter/X users based on the topic. This study uses Google Scholar and the BerTopic algorithm for effective topic modeling on Twitter/X. The model considers unrelated data management, dynamic user credibility, and organizing activities based on the Big Data lifecycle. Finally, using Linear Regression, Support Vector Regression, K-Nearest Neighbor, Random Forest, Classification and Regression Trees algorithms, the model predicts the credibility of Twitter/X users and proves that it performed better than similar models through Classification and Regression Trees. In addition, the model is generalizable for all organizational purposes due to the integration of heterogeneous resources and feature selection methods.

The simplification of the symptom Checklist-90 scale utilizing machine learning techniques

The Symptom Checklist-90 (SCL-90), widely utilized for psychological assessments, faces challenges due to its extensive nature. Streamlining the SCL-90 is essential in order to enhance its practicality without compromising its broad applicability across diverse settings. The objective of this study is to employ machine learning techniques to simplify the dimensions and individual items within each dimension, while simultaneously validating the accuracy and practicality of the streamlined SCL-90 scale. A total of 23,028 valid responses of the SCL-90 were obtained from university students, with positive cases accounting for 49.58 % and negative cases accounting for 50.42 %. The findings demonstrate that by utilizing the Support Vector Classification (SVC) algorithm, it is possible to reduce the scale from ten dimensions to four, achieving an overall prediction accuracy of 89.50 % for the total score. Further simplification of these remaining four dimensions resulted in a reduction from 44 to 29 items per dimension, yielding individual dimension accuracies exceeding 90 %, along with sensitivity and specificity levels surpassing 85 %, and the reliability coefficients consistently exceeded 0.8 across different algorithms. In conclusion, we successfully reduced the number of scale items from 90 to 29, resulting in a reduction of 67.78 % in overall assessment time while maintaining a high reliability coefficient of 0.95. Importantly, the streamlined scale demonstrated no significant decrease in assessment effectiveness. This refined version facilitates rapid comprehension of individuals' comprehensive mental health status and is well-suited for widespread application in experiential settings.

Respective contribution of baseline clinical data, tumour metabolism and tumour blood-flow in predicting pCR after neoadjuvant chemotherapy in HER2 and Triple Negative breast cancer

Background:The aim of this study is to investigate the added value of combining tumour blood flow (BF) and metabolism parameters, including texture features, with clinical parameters to predict, at baseline, the pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) in patients with newly diagnosed breast cancer (BC).Methods:One hundred and twenty-eight BC patients underwent a 18F-FDG PET/CT before any treatment. Tumour BF and metabolism parameters were extracted from first-pass dynamic and delayed PET images, respectively. Standard and texture features were extracted from BF and metabolic images. Prediction of pCR was performed using logistic regression, random forest and support vector classification algorithms. Models were built using clinical (C), clinical and metabolic (C+M) and clinical, metabolic and tumour BF (C+M+BF) information combined. Algorithms were trained on 80% of the dataset and tested on the remaining 20%. Univariate and multivariate features selections were carried out on the training dataset. A total of 50 shuffle splits were performed. The analysis was carried out on the whole dataset (HER2 and Triple Negative (TN)), and separately in HER2 (N=76) and TN (N=52) tumours.Results:In the whole dataset, the highest classification performances were observed for C+M models, significantly (p-value<0.01) higher than C models and better than C+M+BF models (mean balanced accuracy of 0.66, 0.61, and 0.64 respectively). For HER2 tumours, equal performances were noted for C and C+M models, with performances higher than C+M+BF models (mean balanced accuracy of 0.64, and 0.61 respectively). Regarding TN tumours, the best classification results were reported for C+M models, with better performances than C and C+M+BF models but not significantly (mean balanced accuracy of 0.65, 0.63, and 0.62 respectively).Conclusion:Baseline clinical data combined with global and texture tumour metabolism parameters assessed by 18F-FDG PET/CT provide a better prediction of pCR after NAC in patients with BC compared to clinical parameters alone for TN, and HER2 and TN tumours together. In contrast, adding BF parameters to the models did not improve prediction, regardless of the tumour subgroup analysed.

Steady state stability assessment of voltage source converter in a DC microgrid under weak grid conditions

Traditionally, steady-state assessment involves analyzing numerous variables using Eigen analysis. This paper presents a decision support application for diagnosing the steady-state assessment of droop-controlled voltage source inverters in islanded microgrid operations or weak grid operations with reduced input attributes. This paper proposes an approach using feature extraction from the state space variables of the droop-controlled voltage source inverter (VSI). Photovoltaic (PV) and wind energy sources are considered with their stipulated power-delivering capability considered. To improve the generalization of the predictive model, preprocessing techniques are employed to eliminate data distortions. Dimensionality reduction is achieved through principal component analysis (PCA) applied to the steady-state variables. The evaluation of the VSI's steady-state stability is conducted utilizing support vector classification algorithm. To ascertain the reliability of the steady-state stability classification, an assessment of the support vector machine (SVM) model's performance is carried out, which includes the examination of metrics like the area under the curve (AUC) and the receiver operating characteristics (ROC) curve. The findings from the assessment of VSI's steady-state stability indicate a commendable level of performance, achieving an accuracy rate of 93.5%.

Numerical simulation and experimental investigation of pressure pulsation-induced nonlinear characteristics in marine twin-screw pumps under various clearance between rotor and stator

The twin-screw pump (TSP) is widely used in naval engineering for seawater transportation and treatment. However, the diverse composition of seawater, often containing impurities, poses a risk of screw corrosion and jamming of clearances. Detecting the compact internal structure's clearance between rotor and stator (GAPR) proves challenging. To address these challenges, a novel approach combining chaos theory's reconstructive phase space technique was proposed for processing pressure pulsation signals. The research found that only specific monitoring points displayed chaotic characteristics in pressure pulsation signals. The attractor structure complexity increased with GAPR changes, while the chaotic features decreased. Nonlinear analysis of pulsation signals at different GAPR values allowed determining reasonable ranges. Applying support vector classification algorithm based on chaotic dynamics achieved an impressive 89% accuracy in identifying GAPR values. This study offers practical insight for TSP fault detection and operational optimization, holding significance in both theory and practice.

Automated Identification of Ectatomma edentatum (Hymenoptera: Formicidae) using Supervised Algorithms

Taxonomy constantly seeks alternatives to simplify and enhance the identification of living organisms. This study focuses on developing new tools for identifying ant species, aiming to address gaps in determining certain species that pose challenges for naming and study. Species identification can often be a time-consuming and intricate process. We aim to automate the identification process of Ectatomma edentatum (Roger, 1863), utilizing Machine Learning techniques to assess if efficiency can be improved and gaps in ant taxonomy reduced. We applied k-nearest neighbors (KNN) and Support Vector Classification (SVC) algorithms. Adapting these models to the dataset yielded excellent results, with both models demonstrating positive performance in classifying the ants. SVC achieved 100% accuracy, while KNN achieved 96%, affirming the effectiveness of these methods in ant identification. This study highlights the value of supervised algorithms in myrmecology, offering a valuable tool for taxonomy and species classification, ultimately providing accurate synthesis and prediction for species naming.

Comparative Study of Machine Learning Algorithms for Performing Ham or Spam Classification in SMS

Purpose: Fraud is rampant in the current era, especially in the era of technology where there is now easy access to a lot of information. Therefore, everyone needs to be able to sort out whether the information received is the right information or information that is fraudulent. In this research, the process of classifying messages including ham or spam has been carried out. The purpose of this research is to be able to build a model that can help classify messages. The purpose of this research is also to determine which machine learning method can accurately and efficiently perform the ham or spam classification process on messages.Methods: In this research, the ham or spam classification process has been using machine learning methods. The machine learning methods used are the classification process with Random Forest, Logistic Regression, Support Vector Classification, Gradient Boosting, and XGBoost Classifier algorithms. Results: The results obtained after testing in this study are the classification process using the Random Forest algorithm getting an accuracy of 97.28%, Logistic Regression getting an accuracy of 94.67%, with Support Vector Classification getting an accuracy of 97.93%, and using XGBoost Classifier getting an accuracy of 96.47%. The best precision value obtained in this study is 98% when using the random forest algorithm. The best recall value is 94% when using the SVC algorithm. While the best f1-score value is 95% when using the SVC algorithm.Novelty: This research has been compared with several algorithms. In previous research, it is still very rarely done using XGBoost to classify the ham or spam in messages. We focus on giving brief information based con comparison algorithm and show the best algorithm to classify classify the ham or spam in messages. And for the novelty that exists from this research, the machine learning model built gets better accuracy when compared to previous research.

Film Review Sentiment Analysis: Comparison of Logistic Regression and Support Vector Classification Performance Based on TF-IDF

Film sentiment analysis is a process for evaluating a sentiment value that exists in film reviews, so that positive or negative responses from films can be identified. In this study, a sentiment analysis will be carried out on film reviews on IMBD. The analysis was carried out to find out which reviews were positive and negative from film critics. The method used to carry out sentiment analysis in this study is review analysis and processing with TF-IDF and a positive or negative prediction process based on reviews that have been processed using a logistic regression algorithm and support vector classification. The data to be used is film reviews on IMBD, which consists of 2000 data, which is divided into 1000 positive data and 1000 negative data. Which is where the data will be preprocessed first and split with a percentage of 70% training data and 30% testing data. In the prediction process using the logistic regression algorithm, obtaining a test accuracy of 80.61%. While the prediction process using the support vector classification algorithm obtains a test accuracy of 82.42%.

A novel approach for standardizing clinical laboratory categorical test results using machine learning and string distance similarity

Standardizing clinical laboratory test results is critical for conducting clinical data science research and analysis. However, standardized data processing tools and guidelines are inadequate. In this paper, a novel approach for standardizing categorical test results based on supervised machine learning and the Jaro-Winkler similarity algorithm is proposed. A supervised machine learning model is used in this approach for scalable categorization of the test results into predefined groups or clusters, while Jaro-Winkler similarity is used to map text terms into standard clinical terms within these corresponding groups. The proposed method is applied to 75062 test results from two private hospitals in Bangladesh. The Support Vector Classification algorithm with a linear kernel has a classification accuracy of 98%, which is better than the Random Forest algorithm when categorizing test results. The experiment results show that Jaro-Winkler similarity achieves a remarkable 99.93% success rate in the test result standardization for the majority of groups with manual validation. The proposed method outperforms previous studies that concentrated on standardizing test results using rule-based classifiers on a smaller number of groups and distance similarities such as Cosine similarity or Levenshtein distance. Furthermore, when applied to the publicly available MIMIC-III dataset, our approach also performs excellently. All these findings show that the proposed standardization technique can be very beneficial for clinical big data research, particularly for national clinical research data hubs in low- and middle-income countries.

K-outlier removal based on contextual label information and cluster purity for continuous data classification

Outlier detection is extensively adopted in machine learning applications to identify rare but significant objects in a data distribution that deviate from the majority of objects. However, none of the existing outlier definitions use the available contextual label information in a dataset to improve the identification of outliers, though it could improve the performance of data analysis. In this study, we propose a novel definition for outliers considering the available contextual label information along with a method to remove outliers based on the proposed definition to improve classification performance and granulation purity. The experimental results on eight public datasets show that the removal of outliers using the novel method highly improves the classification accuracy with Support Vector Machine, k-Nearest Neighbors, and Classification and Regression Tree algorithms while identifying 100% pure subclasses of the labeled major classes. This method is beneficial in identifying the outliers automatically when the data contains classification information about a specific application instead of outlierness information.

Research on Modeling Weighted Average Temperature Based on the Machine Learning Algorithms

In response to the nonlinear fitting difficulty of the traditional weighted average temperature (Tm) modeling, this paper proposed four machine learning (ML)-based Tm models. Based on the seven radiosondes in the Yangtze River Delta region from 2014 to 2019, four forecasting ML-based Tm models were constructed using Light Gradient Boosting Machine (LightGBM), Support Vector Machine (SVM), Random Forest (RF), and Classification and Regression Tree (CART) algorithms. The surface temperature (Ts), water vapor pressure (Es), and atmospheric pressure (Ps) were identified as crucial influencing factors after analyzing their correlations to the Tm. The ML-based Tm models were trained using seven radiosondes from 2014 to 2018. Then, the mean bias and root mean square error (RMSE) of the 2019 dataset were used to evaluate the accuracy of the ML-based Tm models. Experimental results show that the overall accuracy of the LightGBM-based Tm model is superior to the SVM, CART, and RF-based Tm models under different temporal variations. The mean RMSE of the daily LightGBM-based Tm model is reduced by 0.07 K, 0.04 K, and 0.13 K compared to the other three ML-based models, respectively. The mean RMSE of the monthly LightGBM-based Tm model is reduced by 0.09 K, 0.04 K, and 0.11 K, respectively. The mean RMSE of the quarterly LightGBM-based Tm model is reduced by 0.09 K, 0.04 K, and 0.11 K, respectively. The mean bias of the LightGBM-based Tm model is also smaller than that of the other ML-based Tm models. Therefore, the LightGBM-based Tm model can provide more accurate Tm and is more suitable for obtaining GNSS precipitable water vapor in the Yangtze River Delta region.

Land use change detection using the intensity analysis at the Bosomtwe District, Ashanti Region, Ghana

Changing environmental conditions owing to increasing resource degradation and urbanization in the Bosomtwe district of Ghana make the land cover and change detection analysis an extremely important consideration for Sustainable Environmental Management. The district, where lies the only natural occurring Lake in Ghana, Lake Bosomtwe, is a highly populated urban district that is threatened by rapid vegetation losses, subsequently destroying important biodiversity within the watershed of the UNESCO biosphere reserve. This research investigates the significance of transitions among the major land use/land cover categories in the Bosomtwe district using GIS and remote sensing technologies. Support Vector classification algorithm was adopted for classifying satellite imageries of the years 2002, 2008, 2016, and 2022 while the analysis of the dynamics of change was conducted using the land use change intensity analysis. The study results provide evidence of the prominence of LULC changes within the 20 years interval of assessment. Significant changes in the structure and arrangement of the LULC within the Bosomtwe area were observed for 20 year period. The period between 2016 and 2022 experienced rapid changes and the land use transformation accelerated over the whole period as compared to the 2002-2008- and 2008-2016- year intervals. The study concludes that the major changes are Built-up and Agricultural activities which have claimed forest lands around the Lake's catchment and has intensified in recent years (2016 – 2022). Based on the findings, the study recommends that spatial planning of the spaces in the district be strictly implemented to protect the remnant forest from anthropogenic activities. Afforestation projects should be carried out within the district to reclaim the lost biodiversity. Implementation of such projects are critical in conserving the Lake's biosphere and ecosystem to ensure its sustainability.

Detection of Water Hyacinth (Eichhornia crassipes) in Lake Tana, Ethiopia, Using Machine Learning Algorithms

Lake Tana is Ethiopia’s largest lake and is infested with invasive water hyacinth (E. crassipes), which endangers the lake’s biodiversity and habitat. Using appropriate remote sensing detection methods and determining the seasonal distribution of the weed is important for decision-making, water resource management, and environmental protection. As the demand for the reliable estimation of E. crassipes mapping from satellite data grows, comparing the performance of different machine learning algorithms could help in identifying the most effective method for E. crassipes detection in the lake. Therefore, this study aimed to examine the ability of random forest (RF), support vector machine (SVM), and classification and regression tree (CART) machine learning algorithms to detect E. crassipes and estimating seasonal spatial coverage of the weed on the Google Earth Engine (GEE) platform using Landsat 8 and Sentinel 2 images. Cloud-masked monthly median composite Landsat 8 and Sentinel 2 data from October 2021 and 2022, January 2022 and 2023, March 2022, and June 2022 were used to represent autumn, winter, spring, and summer, respectively. Four spectral indices were derived and used in combination with spectral bands to improve the E. crassipes detection accuracy. All methods achieved greater than 95% and 90% overall accuracy when using Sentinel 2 and Landsat 8 images, respectively. Using both data sets, all methods achieved a greater than 93% F1 score for E. crassipes detection. Though the difference in performance between the methods was small, the RF was the most accurate, while the SVM and CART methods had the same accuracy. The maximum E. crassipes coverage area was observed in autumn (22.4 km2), while the minimum (2.2 km2) was observed in summer. Based on Sentinel 2 data, the E. crassipes area coverage decreased significantly by 62.5% from winter to spring and increased significantly by 81.7% from summer to autumn. The findings suggested that the RF classifier was the most accurate E. crassipes detection algorithm, and autumn was an appropriate season for E. crassipes detection in Lake Tana.

Pulse Bag Filter Fault Classifier Research and Analysis

Aiming at the problem of fault alarm flooding, a fault classifier is established to automatically classify fault types in combination with unsupervised learning methods in machine learning. The k-means method is used to classify the alarm types in dimensionality reduction, and the optimal K value is selected as 11 through various evaluation indexes, and the faults are added to the new category labels according to the center of each cluster after clustering. The random forest and support vector classification algorithm is used to establish a fault classifier model on the fault alarm dataset with new labels, and the unified analysis of the fault type prediction results of the test set unlabeled fault data set is compared with the random forest algorithm, support vector classification, and k-means method. The results show that the unity rate of the prediction results of the three types of classifiers on the test set when taking different K values can reach up to 67%, and the uniformity rate of the prediction results of the three models exceeds 0.998, which has a high degree of uniformity, indicating that the establishment of the fault alarm classifier for the pulse bag dust collector is feasible and has good classification accuracy, which can help maintenance personnel quickly locate and troubleshoot faults, and improve the intelligence level of equipment.

Research on multi-detector real-time fire alarm technology based on signal similarity

Accurate and timely fire detection is the focus of fire prevention, fire control and fire safety. However, fire detectors have a high false alarm rate in structures. To effectively realize real-time fire monitoring, a multi-detector fire detection model based on the asynchronous spatiotemporal signal similarity of different detectors was developed in this study. Firstly, the correlation coefficients of prediction signals and correlation signals were calculated based on the Pearson-Derivative dynamic time warping method. Secondly, a real-time fire detection model based on the signal correlation coefficient was constructed to realize the early detection of fire warnings. Besides, a calculation rule of the correlation coefficient for multi-detector signals was proposed, and the alarm threshold values in fire detection under different numbers of responding detectors were determined by the Support Vector Classification (SVC) algorithm. Finally, the effectiveness of the proposed fire detection model was verified by the fire experiment data from the National Institute of Standards and Technology (NIST) and the kitchen cooking nuisance tests.

Cellular Automata-Based Artificial Neural Network Model for Assessing Past, Present, and Future Land Use/Land Cover Dynamics

Land use and land cover change (LULCC) is among the most apparent natural landscape processes impacted by anthropogenic activities, particularly in fast-growing regions. In India, at present, due to the impacts of anthropogenic climate change, supplemented by the fast pace of developmental activities, the areas providing the highest agricultural yields are facing the threat of either extinction or change in land use. This study assesses the LULCC in the fastest-changing landscape region of the Indian state of Bihar, District Muzaffarpur. This district is known for its litchi cultivation, which, over the last few years, has been observed to be increasing in acreage at the behest of a decrease in natural vegetation. In this study, we aim to assess the past, present and future changes in LULC of the Muzaffarpur district using support vector classification and CA-ANN (cellular automata-artificial neural network) algorithms. For assessing the present and past LULC of the study area, we used Landsat Satellite data for 1990, 2000, 2010, and 2020. It was observed that between 1990 and 2020, the area under vegetation, wetlands, water body, and fallow land decreased by 44.28%, 34.82%, 25.56%, and 5.63%, respectively. At the same time, the area under built-up, litchi plantation, and cropland increased by 1451.30%, 181.91%, and 5.66%, respectively. Extensive ground truthing was carried out to assess the accuracy of the LULC for 2020, whereas historical google earth images were used for 1990, 2000, and 2010, through the use of overall accuracy and kappa coefficient indices. The kappa coefficients for the final LULC for the years 1990, 2000, 2010, and 2020 were 0.79, 0.75, 0.87, and 0.85, respectively. For forecasting the future LULC, first, the LULC of 1990 and 2010 were used to predict the landscape for 2020 using the CA-ANN model. After calibrating and validating the CA-ANN outputs, LULC for 2030 and 2050 were generated. The generated future LULC scenarios were validated using kappa index statistics by comparing the forecast outcomes with the original LULC data for 2020. It was observed that in both 2030 and 2050, built-up and vegetation would be the major transitioning LULC. In 2030 and 2050, built-up will increase by 13.15% and 108.69%, respectively, compared to its area in 2020; whereas vegetation is expected to decrease by 14.30% in 2030 and 32.84% in 2050 compared to its area in 2020. Overall, this study depicted a decline in the natural landscape and a sudden increase in the built-up and cash-crop area. If such trends continue, the future scenario of LULC will also demonstrate the same pattern. This study will help formulate better land use management policy in the study area, and the overall state of Bihar, which is considered to be the poorest state of India and the most vulnerable to natural calamities. It also demonstrates the ability of the CA-ANN model to forecast future events and comprehend spatiotemporal LULC dynamics.

A Filter-Based Feature-Engineering-Assisted SVC Fault Classification for SCIM at Minor-Load Conditions

In most manufacturing industries, squirrel cage induction motors (SCIMs) are essential due to their robust nature, high torque generation, and low maintenance costs, so their failure often times affects productivity, profitability, reliability, etc. While various research studies presented techniques for addressing most of these machines’ prevailing issues, fault detection in cases of low slip or, low load, and no loading conditions for motor current signature analysis still remains a great concern. When compared to the impact on the machine at full load conditions, fault detection at low load conditions helps mitigate the impact of the damage on SCIM and reduces maintenance costs. Using stator current data from the SCIM’s direct online starter method, this study presents a feature engineering-aided fault classification method for SCIM at minor-load conditions based on a filter approach using the support vector classification (SVC) algorithm as the classifier. This method leverages the loop-hole of the Fourier Transform at minor-load conditions by harnessing the uniqueness of the Hilbert Transform (HT) to present a methodology that combines different feature engineering technologies to excite, extract, and select 10 discriminant information using a filter-based approach as the selection tool for fault classification. With the selected features, the SVC performed exceptionally well, with a significant diagnostic performance accuracy of 97.32%. Further testing with other well-known robust classifiers such as decision tree (DT), random forest (RF), k-nearest neighbor (KNN), gradient boost classifier (GBC), stochastic gradient descent (SGD), and global assessment metrics revealed that the SVC is reliable in terms of accuracy and computation speeds.

Forest Tree Species Classification Based on Sentinel-2 Images and Auxiliary Data

Most research on forest tree species classification based on optical image data uses information such as spectral reflectance, vegetation index, texture, and phenology data. However, owing to the limited spectral resolution of multispectral images and the high cost of hyperspectral data, there is room for improvement in the classification of tree species in large areas based on optical images. The combined application of multispectral images and other auxiliary data can provide a new method for improving tree species classification accuracy. Hence, Sentinel-2 images were used to extract spectral reflectance, spectral index, texture, and phenological information. Data for topography, precipitation, air temperature, ultraviolet aerosol index, NO2 concentration, and other variables were included as auxiliary data. Models for forest tree species classification were constructed through feature combination and feature optimization using the random forest (RF), gradient tree boost (GTB), support vector machine (SVM), and classification and regression tree (CART) algorithms. The classification results of 16 feature combinations with the 4 classification methods were compared, and the contributions of different features to the classification models of forest tree species were evaluated. Finally, the optimal classification model was selected to identify the spatial distribution of forest tree species in the study area. The model based on feature optimization gave the best results among the 16 feature combination models. The overall accuracy and kappa coefficient were increased by 18% and 0.21, respectively, compared with the spectral classification model, and by 17% and 0.20, respectively, compared with the spectral and spectral index classification model. By analyzing the feature optimization model, it was found that terrain, ultraviolet aerosol index, and phenological information ranked as the top three features in terms of importance. Although the importance of spectral reflectance and spectral index features was lower, the number of feature variables accounted for a large proportion of the total. The importance of commonly used texture features was limited, and these features were not present in the feature optimization model. The RF algorithm had the highest classification accuracy, with an overall accuracy of 82.69% and a kappa coefficient of 0.80, among the four classification algorithms. The results of GTB were close to those of RF, and the difference in overall classification accuracy was only 0.14%. However, the results of the SVM and CART algorithms were relatively weaker, with overall classification accuracies of about 70%. It can be concluded that the combined application of Sentinel-2 images and auxiliary data can improve forest tree species classification accuracy. The model based on feature optimization achieved the highest classification accuracy among the 16 feature combination models. The spectral reflectance and spectral index data extracted from optical images are useful for tree species classification, but the effect of texture features was very limited. Auxiliary data, such as topographic features, ultraviolet aerosol index, phenological features, NO2 concentration features, topographic diversity features, precipitation features, temperature features, and multi-scale topographic location index data, can effectively improve forest tree species classification accuracy. The RF algorithm had the highest accuracy, and it can be used for tree species classification space distribution identification. The combined application of Sentinel-2 images and auxiliary data can improve classification accuracy, but the highest accuracy of the model was only 82.69%, which leaves room for improvement. Thus, more effective auxiliary data and the vertical structural parameters extracted from satellite LiDAR can be combined with multispectral images to improve forest tree species classification accuracy in future research.

Development of a Machine Learning-Based Screening Method for Thyroid Nodules Classification by Solving the Imbalance Challenge in Thyroid Nodules Data.

This study aims to show the impact of imbalanced data and the typical evaluation methods in developing and misleading assessments of machine learning-based models for preoperative thyroid nodules screening. A retrospective study. The ultrasonography features for 431 thyroid nodules cases were extracted from medical records of 313 patients in Babol, Iran. Since thyroid nodules are commonly benign, the relevant data are usually unbalanced in classes. It can lead to the bias of learning models toward the majority class. To solve it, a hybrid resampling method called the Smote-was used to creating balance data. Following that, the support vector classification (SVC) algorithm was trained by balance and unbalanced datasets as Models 2 and 3, respectively, in Python language programming. Their performance was then compared with the logistic regression model as Model 1 that fitted traditionally. The prevalence of malignant nodules was obtained at 14% (n = 61). In addition, 87% of the patients in this study were women. However, there was no difference in the prevalence of malignancy for gender. Furthermore, the accuracy, area under the curve, and geometric mean values were estimated at 92.1%, 93.2%, and 76.8% for Model 1, 91.3%, 93%, and 77.6% for Model 2, and finally, 91%, 92.6% and 84.2% for Model 3, respectively. Similarly, the results identified Micro calcification, Taller than wide shape, as well as lack of ISO and hyperechogenicity features as the most effective malignant variables. Paying attention to data challenges, such as data imbalances, and using proper criteria measures can improve the performance of machine learning models for preoperative thyroid nodules screening.