Conventional Classification Algorithms Research Articles

Smart grid security enhancement by detection and classification of non‐technical losses employing deep learning algorithm

Non-technical loss (NTL) is detrimental to the smart grid. Intelligent application of advanced metering infrastructure (AMI) helps to solve NTL detection and classification. By using advanced learning algorithms, data analysis on the massive data generated by AMI is helpful in the detection and classification of electricity theft. Conventional data analysis algorithm, like Support Vector Machine (SVM), Random Forest Algorithm (RFA), and 1D- Conventional Neural Network (1D-CNN), has low detection and classification accuracy of electricity theft. Because these methods failed to predict and classify multidimensional electricity consumption data by various consumers in AMI in the smart grid. In this research work, a multidimensional deep learning algorithm is proposed to learn and classify the non-periodicity of electricity. This helps to detect electricity theft by a consumer from the periodic load curve. Both weekly load patterns and daily load patterns are processed as 2D electricity data samples. From the proposed multidimensional deep learning model, an average classification accuracy of 97.5% and a precision-recall of 0.97 were obtained. This validates that the proposed deep learning model outperforms other conventional data analysis classification algorithm.

Clinical questionnaire filling based on question answering framework

BackgroundElectronic Health Records (EHR) are the foundation of much medical research. However, analyzing the text data of EHRs directly is an challenging task. Therefore, physicians often use questionnaires to first convert text data to structured data. Filling in these questionnaires requires a considerable amount of time and medical knowledge. It is a significant task to develop an algorithm to make computers fill out these questionnaires automatically. ObjectiveThis research aims to build a deep learning model that can automatically complete questionnaires with given medical text. MethodsThis task is a part of Information Extraction (IE), but it differs from the existing tasks of medical IE. Because of the questions in questionnaires are closed-end type, which refers to making a selection among given options, we could treat this task as a classification problem. However, conventional classification algorithms are resource-consuming when filling out one entire questionnaire with one model. They also could not use the question information to guide the questionnaire filling task. To handle these issues, we propose a neural network model based on question answering (QA) framework in this paper. With this framework, our neural network can fill out one complete questionnaire using only one model. ResultsWe perform experiments on three real-world Chinese medical datasets and related clinical questionnaires. Our model respectively achieves F1 scores of 0.9273, 0.8834, and 0.9846. The results outperform several baseline models. ConclusionThe strong performance of our QA model will allow us to build a system which can help physicians to fill out questionnaires and convert text data to structured data. This system can reduce the workload of physicians.

Ship Classification from SAR Images based on Sequence Input of Deep Neural Network

This article proposes an all brand new classification architecture for SAR images of the ship via deep learning. Compared with the most widely used conventional classification algorithm based on deep neural networks, we set a sequence of SAR images as input to the neural network, which is different from a single image as input in those kinds of conventional ones. Two central neural networks form the construction of this new classification architecture; the first is a convolutional neural network used to extract the features of each frame, and the second is the LSTM part, which is trained on the sequence to predict the labels. It is much more significant to enhance the connection among the SAR images primarily due to the limit of SAR data, which makes us combine all the SAR images to several sequences. The proposed new classification architecture is trained on the OpenSARShip[2] dataset captured by Sentinel-1 space-borne radar. Experiment results show that the classification architecture can get an accuracy of 99.24% in classifying the six kinds of targets, which is much more precise than any conventional methods before.

Correction to: Computer-aided diagnosis (CAD) system based on multi-layer feature fusion network for skin lesion recognition in dermoscopy images

Skin lesion recognition is one of the most important tasks in dermoscopic image analysis. Current Convolutional Neural Network (CNN) algorithms based recognition methods tend to become a standard methodology to fix a large array of Computer-Aided Diagnosis (CAD) and interpretation problems. Besides significant practical and theoretical improvements in their architecture, their effectiveness is built on the existence of the flexible pre-trained models which generalize well to novel tasks and handle the problem of having small set of dermoscopic data. However, existing works pay little attention to exploring the benefits of hierarchical multi-feature fusion for classifying the skin lesions in digital dermoscopic images. Practically, it has been found that integrating multi-layer features has significant potential for improving performance of any pattern recognition task. In this paper, we developed a robust CAD system based on transfer learning and multi-layer feature fusion network to diagnose complex skin diseases. It is a convenient approach in terms of overfitting prevention, convergence speed and high morphological feature similarity processing. Our research focuses exclusively on obtaining optimal performance with addressing the various gaps in the skin pattern recognition area. For validation and comparison purposes, the proposed approach was evaluated on publicly dermoscopic dataset, and achieved the high recognition precision compared with fully trained CNN models, fine-tuning process, single CNN model and other related works. Therefore, the study demonstrates that our proposed approach can dramatically improve the performance of CAD systems which are based on the conventional recognition and classification algorithms for skin lesion recognition in dermoscopic data.

Towards an End-to-End Method for High-Variance SERS Spectra Classification and Quantification

Surface enhanced Raman spectroscopy (SERS) is a subfield of Raman spectroscopy where trace compounds can be detected with the help of plasmonic materials. They enhance the Raman signal from molecules in proximity. Our work exploits the low number of molecules with enhanced signal by taking sequences of short integration spectra. This enable us to see events where a Raman active molecule shortly adsorb to the surface in an enhancing spot and gives us strong, but anisotropic signal. Because of this, conventional classification algorithms are not suited for efficient classification.The novel data analysis method presented here is based on active spectrum detection, where spectra are kept based on how many Raman peaks they are composed of. The end goal is the characterization and quantification of common drugs and cutting agents in a short time frame. A reusable microfluidic SERS sensor is used in a batch flow way, such as a technician could inject a diluted drug sample in it, analyse it for a few minutes and get result on the screen.The current architecture of the classification of the spectra can be represented with the flowchart as seen on figure 1, where the preprocessing and active spectrum detection are both classical algorithms, in the sense that they are not being optimized by training like machine learning ones, and simply apply the same rules for each and every spectrum. This preprocessing exploits a Savitzky-Golay smoothing, a baseline generation with airPLS, and a peak finding function to select valid peaks and count them and determine if a spectrum is active or not. Afterwards, the active spectra are processed with a multiclass CNN.To create a streamlined quantification pipeline, two quantification algorithms are compared with drug spectra: a CNN that is processing “stitchings” of multiple random spectra and a custom network based on modular neural networks. Both are also compared with various active spectrum detection and classifier methods.The current active spectrum detection is based on peak detection, with an arbitrary threshold set in a manner that if a peak is over its height, it is an active peak, and a spectrum would need an arbitrary number of active peaks to be considered active. This method can sort good spectra from the inactive and weakly active one, but the weak one can be filled with information that could be processed by the classification network. The proposed quantification pipeline forces the next network to be dependent of the previous ones, and such, parallel optimization is possible during training. By setting a quality score based on how active the spectra are, we can let the network optimize the trade-off between how many spectra will be considered for classification and the accuracy of the classification, because weakly active spectra can lower the accuracy when they don’t have enough information to be correctly classified.As mentioned earlier, a multiclass CNN is used. This type of classifier uses a All vs. All (AVA) strategy. This implies that all classes are tested against each other, and that one of the classes must be chosen. In the case of physiological spectra classification, this is problematic. The active spectrum of an unknown molecule would reach the classifier and be forced into one of the learned classes, and the result would be inherently false.To avoid this problem, two different type of classification architecture are tested with the quantification network: 1. An ensemble of binary CNN classifiers will be tested with a One vs. All (OVA) strategy, where there are only 2 outcomes for each classifier.2. An ensemble of one-class classifiers (OCC) where the outcome possible is either Yes or No for one class each. They might look the same at first hand but the difference is vast. This is also called Concept Learning and Anomaly Detection. Figure 1

Type 2 Machine Learning: An Effective Hybrid Prediction Model for Early Type 2 Diabetes Detection

Importance: Diabetes is a chronic disease that can cause long term damage to various parts of the body. To prevent diabetic complications, different attempts integrating machine learning with medicine have been made for building models to predict whether a patient has diabetes or not, but predicting this disease still has room for improvement. Hybrid prediction model presents a novel method and mostly achieve a much better optimal outcome than single classical machine learning algorithms. Objective: To develop a high accuracy model for different onsets of type 2 diabetes prediction. In this way, the integration between clustering and classification techniques can be improved to help detecting diabetes at an earlier stage without deleting observations with missing values and also decrease insignificant features to get the most related features during data collection. Methods: We implement a noise reduction based technique using Kmeans clustering followed by running the Random forest and XGBoost classifiers to extract the unknown hidden features of the dataset and for more accurate results. Results: Prediction accuracy can be observed by benchmarking our model against up-to-date predictive models and common classification algorithms. With an accuracy of 97.53% by 10 fold cross validation, our T2ML model reaches a better accuracy compared with other experiments reported by other researchers in the literature and over various conventional classification algorithms.

Deep reinforcement learning for imbalanced classification

Data in real-world application often exhibit skewed class distribution which poses an intense challenge for machine learning. Conventional classification algorithms are not effective in the case of imbalanced data distribution, and may fail when the data distribution is highly imbalanced. To address this issue, we propose a general imbalanced classification model based on deep reinforcement learning. We formulate the classification problem as a sequential decision-making process and solve it by deep Q-learning network. The agent performs a classification action on one sample at each time step, and the environment evaluates the classification action and returns a reward to the agent. The reward from minority class sample is larger so the agent is more sensitive to the minority class. The agent finally finds an optimal classification policy in imbalanced data under the guidance of specific reward function and beneficial learning environment. Experiments show that our proposed model outperforms the other imbalanced classification algorithms, and it can identify more minority samples and has great classification performance.

Semi-supervised sparse representation classifier (S3RC) with deep features on small sample sized hyperspectral images

Hyperspectral images usually have small number of labeled samples because of the labeling cost and process difficulty. In conventional classification algorithms, classifier performance is depending on the number training samples. In this study, a deep learning based semi-supervised learning framework is proposed to solve this limited labeled sample size problem by utilizing power of labeled and unlabeled samples. The main aim of the study is constructing a general purpose deep model for a specific hyperspectral sensor type and using the model with little effort for all data sets obtained from this sensor type. In proposed framework, the model is trained with a data set from a hyperspectral sensor subsequently it is fine-tuned and evaluated with another data set acquired from the same sensor. Linearly separable deep features of the evaluation data set are extracted from the fine-tuned general deep model. Additionally, a new data formation method is proposed in the transition from hyperspectral data sub-cubes to the deep neural network input. Besides that, three different clustering methods have been used for selecting the initial labeled samples in the semi-supervised learning phase to observe the effects of the sample selection comparatively. As an another contribution of the study, a new semi-supervised sparse representation classifier (S3RC) is proposed with labeled and unlabeled sample information by using linearly separable deep features. The performance of the proposed framework is proven by the experimental results with using small sample sizes.

Highly Accurate Melanoma Detection from Skin Images using Multiple Feature Extraction and Support Vector Machine (SVM)

Digital image processing (DIP) plays a major role in the biomedical image segmentation and classification process. Melanoma detection from skin images is most widely using the application in biomedical imaging. In this paper, a novel algorithm for Melanoma Detection from digital images using multiple feature extraction with a Support Vector Machine (SVM) for highly accurate classification. Multiple features such as color, texture and statistical features are considered to train the SVM. To increase the classification sensitivity, feature extraction is done in (Red, Green, Blue) RGB and (Hue, Intensity, Saturation) HIS color domains. Two separate modules for training and testing is performed with collected sample data by using medical experts. To separate the region from the skin, a cantor-based segmentation technique is used in the gray level component of the input image. The proposed method is tested with various images that are collected from different patients from different places. Form the result validation it is clear that the proposed algorithm can give maximum accuracy of 95% which is best when compared to the conventional classification algorithms

Adaptive weighted over-sampling for imbalanced datasets based on density peaks clustering with heuristic filtering

Learning from imbalanced datasets poses a major challenge in data mining community. When dealing with imbalanced datasets, conventional classification algorithms generally perform poorly as they are originally designed to work under balanced class distribution scenarios. Although there exist different methods to addressing this issue, sampling methods especially over-sampling techniques have shown great potentials as they aim to improve datasets itself rather than the classifiers, which can allow them to be used for any classifier. In this paper, we propose a novel adaptive weighted over-sampling for imbalanced datasets based on density peaks clustering with heuristic filtering. Unlike other clustering-based over-sampling methods, the proposed approach applies modified density peaks clustering rather than traditional k-means clustering techniques to cluster the minority instances due to its capability of accurately identifying sub-clusters with different sizes and densities, which is beneficial for the proposed method to simultaneously accommodate for between-class and within-class imbalance issues caused by various reasons. Subsequently, the size for each identified sub-cluster to be oversampled is adaptively determined according to its own size and density and then the minority instances within each sub-cluster are oversampled based on their probabilities inversely proportional to their distances to the majority class and their densities with the aim of generating more synthetic minority instances for borderline and sparser ones. Finally, in order to avoid the generation of overlapping, a heuristic filtering strategy is also developed to iteratively move the possibly overlapped minority instances away from the majority class. The extensive experimental results on the different imbalanced datasets demonstrate that the proposed approach can achieve better classification performance in most datasets as compared to the other existing over-sampling techniques.

Multi-Modal Data Fusion Using Deep Neural Network for Condition Monitoring of High Voltage Insulator

A novel Fusion Convolutional Network (FCN) is proposed in this research for potential real-time monitoring of insulators using unmanned aerial vehicle (UAV) edge devices. Precise airborne imaging of outdoor objects, such as high voltage insulators, suffers from varied object resolution, cluttered backgrounds, unclear or contaminated surfaces, and illumination conditions. Accurate information about the insulator surface condition is essential and is of a high priority since insulator breakdown is a leading cause of electrical failure. A multi-modal information fusion (MMIF) system is developed during this research to analyze and classify possible contaminations present on the electrical insulators. A novel system, referred to as FCN, consists of a Convolutional Neural Network (CNN) and a binary Multilayer Neural Network (MNN) sub-classifier. While constructing the MMIF dataset for training and testing the novel FCN, the image classification output of the CNN is combined with the leakage current values (LCV) obtained as the classification output of MNN. Each sample of the MMIF dataset is, therefore, represented as a series of fusions. Later, sub-classifiers, of the FCN, are trained to identify the contamination types in the fusion series by implementing a voting system of sub-classifiers which is trained to identify a given class. As a result of the implementation of the proposed FCN, the classification accuracy increased by 8.4%, i.e., from 92% to 99.76%. To compare and benchmark the performance of proposed FCN, conventional classification algorithms are also implemented on the fusion of features that are extracted employing the wavelet transform and PCA methods. State-of-the-art CNN architectures are also discussed on account of their time consumption and memory usage. The conceptualization of a potential hardware implementation of the proposed FCN, on emerging edge devices, is also provided for completeness of the discussion. Pertinent outcomes of this research can be further extended to other potential applications of airborne imaging.

Development and comparative performance evaluation of neural network classification techniques for manufacturing sector firms on the basis of new product growth rate

Continuous changes and improvements in new product development and its process have brought competition to its peak among the manufacturing sector firms. Incessant improvement considers as the only option for sustainable development and growth. The new products have been launched by firms at regular interval of time to stay updated with present market trends and followings. Recognising current status of the firm with respect to its competitors is of utmost importance for better strategy formulation and decision making. In this study electronics, garment and metal and machinery industries are chosen for an in-depth analysis because of enormous growth shown by them in past few years. The firms are categorised into three classes termed as innovative, mediocre and traditional on the basis of their new product growth rate (NPGR). Adaptive exponential functional linked artificial neural network (AEFLANN) along with other neural network models and conventional methods of classification has been employed. An extension of AEFLANN model using derivative-based (RLS) and derivative free (PSO) algorithm has been proposed. Comparative performance analysis for neural network and conventional classification algorithms has been made. Neural network classification models outperform other employed models in terms of classification accuracy.

Electroencephalogram-Based Preference Prediction Using Deep Transfer Learning

Transfer learning is an approach in machine learning where a model that was built and trained on one task is re-purposed on a second task. The success of transfer learning in computer vision has motivated its use in neuroscience. Although common in image recognition, the use of transfer learning in EEG classification remains unexplored. Most EEG-based neuroscience studies depend on using traditional machine learning algorithms to answer a question, rather than on improving the algorithms. Developing algorithms for transfer learning for EEG can also assist with problems of low data availability in EEG classification. The primary objective of this study is to investigate EEG-based transfer learning and propose deep transfer learning models to transfer knowledge from emotion recognition to preference recognition to enhance the classification prediction accuracy. To the best of our knowledge, this is the first study demonstrating the effect of applying deep transfer learning between EEG-based emotion recognition and EEG-based preference detection. We propose different approaches for deep transfer learning models to detect preferences from EEG signals using the preprocessed DEAP dataset. Two types of features were extracted from EEG signals, namely the power spectral density and valence. We built three models of deep neural networks: basic without transfer learning, fine-tuning of deep transfer learning, and retraining of deep transfer learning. We compared the performance of deep transfer learning with those of deep neural networks and other conventional classification algorithms such as support vector machine, random forest, and k-nearest neighbor. Although the deep neural network classifiers achieved a high accuracy of greater than 87%, deep transfer learning achieved the highest accuracy result of 93%. The results demonstrate that although the proposed deep transfer learning approaches exhibit higher accuracy than the support vector machine and k-nearest neighbor classifiers, random forest achieves results similar to those of deep transfer learning.

Data Fusion Generative Adversarial Network for Multi-Class Imbalanced Fault Diagnosis of Rotating Machinery

For the fault diagnosis problems of rotating machinery in the real industrial practice, measurement data with imbalanced class distributions negatively affect the diagnostic performance of most conventional machine learning classification algorithms since equal cost weights are assigned to different fault classes. Meanwhile, the widely used traditional data generation methods for the imbalanced data problem are limited by data dependencies over time continuity. To fill this research gap, this paper develops a new diagnostic framework based on the adversarial neural networks (GAN) and multi-sensor data fusion technique to generate new synthetic data for data compensation purpose. Two different practice modes are designed based on this framework according to the position logic of the data fusion, namely a Pre-fusion GAN mode and a Post-fusion GAN mode. More concretely, without data pre-processing, the designed generator generates synthetic data to puzzle the discriminator and the synthetic data that out-trick the discriminator can be used to compensate the minor class. To avoid data dependency and to ensure the generality of the proposed framework, the network modelling are trained with a more practical approach where the training and test data are obtained under different rotating speeds. Two imbalanced data sets on the rotating machinery, one benchmark public rolling bearing data set and another gear box data set acquired in our lab, are used to validate the proposed method. The performance is examined through a wide range of data imbalanced ratios (as high as 30:1), and compared with other state-of-the-art methods. The experiment results conclude that the proposed Pre-fusion GAN and Post-fusion GAN frameworks both have good performance on the imbalanced fault diagnosis of rotating machinery.

Predicting Outcomes of Legal Cases based on Legal Factors using Classifiers

Predicting outcomes of legal cases may aid in the understanding of the judicial decision-making process. Outcomes can be predicted based on i) case-specific legal factors such as type of evidence ii) extra-legal factors such as the ideological direction of the court. The details of case-specific legal factors can be extracted from legal judgments. However, extracting these factors from legal texts is a tedious and time-consuming process. In this work, important factors affecting outcomes of murder related cases (taken from Delhi District Court) are identified and a database of 86 cases is prepared in order to use these factors as descriptors for outcome prediction. The outcome prediction is seen as a binary classification problem for classes ‘Acquittal’ and ‘Conviction’ of the accused person. Conventional machine learning classification algorithms are applied and Leave-one-out cross validation is used to produce the results. The performance of classifiers is evaluated and compared using metrics such as Accuracy, Precision, Recall, and F1 Score. The statistical distribution of features and the experimental results (Accuracy ranging from 85% to 92% and F1 Score from 86% to 92% for classifiers) show the success in identifying important factors of concerned cases and in turn predicting their outcomes.

VALIDATION ASSESSMENTS ON RESAMPLING METHOD IN IMBALANCED BINARY CLASSIFICATION FOR LINEAR DISCRIMINANT ANALYSIS

The curse of class imbalance affects the performance of many conventional classification algorithms including linear discriminant analysis (LDA). The data pre-processing approach through some resampling methods such as random oversampling (ROS) and random undersampling (RUS) is one of the treatments to alleviate such curse. Previous studies have attempted to address the effect of a resampling method on the performance of LDA. However, some studies contradicted with each other based on different performance measures as well as validation strategies. This manuscript attempted to shed more light on the effect of a resampling method (ROS or RUS) on the performance of LDA based on true positive rate and true negative rate through five validation strategies, i.e. leave-one-out cross-validation, k-fold cross-validation, repeated k-fold cross-validation, naive bootstrap, and .632+ bootstrap. 100 two-group bivariate normally distributed simulated and four real data sets with severe class imbalance ratio were utilised. The analysis on the location and dispersion statistics of the performance measures was further enlightened on: (i) the effect of a resampling method on the performance of LDA, and (ii) the enhancement in the learning fairness of LDA on objects regardless of sample size, hence reducing the effect of the curse of class imbalance.

Enhanced Classifier Accuracy in Liver Disease Diagnosis using a Novel Multi Layer Feed Forward Deep Neural Network

Classification techniques are often used for predicting Liver diseases and assist doctors in early detection of liver diseases. As per studies in the past and our experiments, conventional classification algorithms are found to be less accurate in predicting liver diseases. Therefore, there is a need for sophisticated classifiers in this area. For many medical applications, including Liver Diseases, Deep Neural Networks (DNNs) are used but the accuracies are not satisfactory. Deep Neural Network training is a time taking procedure, particularly if the hidden layers and nodes are more. Most of the times it leads to over fitting and the classifier does not perform well on unseen data samples .We, in this paper, tuned a Multi Layer Feed Forward Deep Neural Network (MLFFDNN) by fitting appropriate number of hidden layer and nodes, dropout function after each hidden layer to avoid over fitting, loss functions, bias, learning rate and activation functions for more accurate liver disease predictions. We used a balanced data set containing 882 samples. The data is collected from north coastal districts of Andhra Pradesh hospitals, India. The training process is carried out for 400 epochs and finally It is .observed that our model exhibited 98% accuracy at epoch 363 which is more than the performance of Neural Network models tuned till now by machine learning researchers and also some regularly used classification algorithms like Support Vector Machines (SVM), Naive Bayes (NB), C4.5 Decision Tree, Random Belief Networks and Alternating Decision Trees (ADT) .

Explorative Study on Mapping Surface Facies of Selected Glaciers from Chandra Basin, Himalaya Using WorldView-2 Data

Mapping of surface glacier facies has been a part of several glaciological applications. The study of glacier facies in the Himalayas has gained momentum in the last decade owing to the implications imposed by these facies on the melt characteristics of the glaciers. Some of the most commonly reported surface facies in the Himalayas are snow, ice, ice mixed debris, and debris. The precision of the techniques used to extract glacier facies is of high importance, as the result of many cryospheric studies and economic reforms rely on it. An assessment of a customized semi-automated protocol against conventional and advanced mapping algorithms for mapping glacier surface facies is presented in this study. Customized spectral index ratios (SIRs) are developed for effective extraction of surface facies using thresholding in an object-based environment. This method was then tested on conventional and advanced classification algorithms for an evaluation of the mapping accuracy for five glaciers located in the Himalayas, using very high-resolution WorldView-2 imagery. The results indicate that the object-based image analysis (OBIA) based semi-automated SIR approach achieved a higher average overall accuracy of 87.33% (κ = 0.85) than the pixel-based image analysis (PBIA) approach. Among the conventional methods, the Maximum Likelihood performed the best, with an overall accuracy of 78.71% (κ = 0.75). The Constrained Energy Minimization, with an overall accuracy of 68.76% (κ = 0.63), was the best performer of the advanced algorithms. The advanced methods greatly underperformed in this study. The proposed SIRs show a promise in the mapping of minor features such as crevasses and in the discrimination between ice-mixed debris and debris. We have efficiently mapped surface glacier facies independently of short-wave infrared bands (SWIR). There is a scope for the transferability of the proposed SIRs and their performance in varying scenarios.

Transmission Line Fault Classification Using Hidden Markov Models

The maintenance of power quality in electrical power systems depends on addressing the major disturbances that may arise during generation, transmission and distribution. Many studies aim to investigate these disturbances by analyzing the behavior of the electrical signal through the classification of short circuit faults in power transmission lines as a way to assist the administration and maintenance of the electrical system. However, most fault classification methods generate a high computational cost that do not always yield satisfactory results; these methods utilize front ends in data processing before being processed by conventional classification algorithms such as Artificial Neural Network (ANN), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Forest (RF) that are adopted into the Frame Based Sequence Classification (FBSC) architecture that uses the front ends Wavelet energy, Waveletconcat, RAW, Root Mean Square front ends (RMS) and Con cat Front End. An alternative method for classifying faults without having to use front ends employs the UFPAFaults database and the Hidden Markov Model (HMM) algorithm that directly treats the electrical signal in the form of multivariate time series. The results indicate the HMM algorithm as a potential classifier because its comparatively low error rate of 0.03% exceeds the performance of the conventional classifiers ANN, SVM, KNN and RF as used with the FBSC architecture. When the statistical test with a significance of α =5 % is applied, only the ANN and RF classifiers present a result close to what the HMM algorithm provides. Another relevant factor is that the HMM algorithm considerably decreases the computational cost by more than 90% of processing time as compared to the conventional classifiers of the FBSC architecture, thereby validating its potential in the direct classification of faults in electric power system transmission lines.

A Classification Method Based on Feature Selection for Imbalanced Data

Imbalanced data are very common in the real world, and it may deteriorate the performance of the conventional classification algorithms. In order to resolve the imbalanced classification problems, we propose an ensemble classification method that combines evolutionary under-sampling and feature selection. We employ the Bootstrap method in original data to generate many sample subsets. $V$ -statistic is developed to measure the distribution of imbalanced data, and it is also taken as the optimization objective of the genetic algorithm for the under-sampling sample subsets. Moreover, we take $F_{1}$ and Gmean indicators as two optimization objectives and employ the multiobjective ant colony optimization algorithm for feature selection of resampled data to construct an ensemble system. Ten low-dimensional and four high-dimensional typical imbalanced datasets are used in experiments. The six state-of-the-art algorithms and four measures are taken for a fair comparison. The experimental results show that our proposed system has a better classification performance compared with other algorithms, especially for the high-dimensional imbalanced data.