Search

Data quality estimation has become an important research axis in applications when defining useful information is required. Determining useful information is an essential task in the learning process. One important aspect thereby is to find optimal and sufficient information that has a good quality and which is complimentary to improve the performance of the decision-making system. In this paper, a survey of data quality is operated in the possibilistic framework. The main objective of this work is the development of a possibilistic approach that aims to extract useful information from imperfect knowledge. The new method consists of two main steps, the classification of feature quality and the determination of the optimal feature subset. The optimal feature subset is deduced according to the feature classification technique and the feature complementarity. To show the performance of the proposed approach, the new method has been evaluated, firstly, on a synthetic dataset, and then on different benchmark databases. Later, the new method is evaluated on two real databases to gain a better insight into the performance of this method. Experiments showed the effectiveness of the proposed method in extracting useful information using the optimal feature subset. Experimental results illustrate significant accuracy rates compared with different earlier methods, especially when the system complexity increases. • This work aims to develop a possibilistic approach for extracting useful data. • Our method involves classifying features based on their quality and defining the optimal subset • Our method is evaluated on synthetic dataset, benchmark databases and two real databases. [ABSTRACT FROM AUTHOR]

• Distance and Shape signatures, individually and combined, appear as new contribution. • Stability indices corresponding to Distance and Shape have been proposed. • Signatures and Stability indices and statistical metrics of relevance form the feature set. • MLP is learnt with this feature set for emotion estimation from face images. • Results obtained thereof are extremely encouraging. Distance and Shape signature features in human faces offer formidable significance in recognizing facial expressions. Identifying appropriate landmarks is a crucial as well as challenging issue in the field of expression recognition of human faces. Appearance model has been found useful to detect the salient landmarks on human faces. These salient landmarks induce a grid on the human face along with the formation of possible triangles joining the grid. Normalized distance and shape signatures are determined from the grid. Distance signature as well as shape signature find respective stability indices which play important role to recognize the facial expressions. Statistical measures such as range, moment, skewness, kurtosis and entropy are calculated from normalized distance and shape signature pair to supplement the feature set. This enhanced feature set is fed into a Multilayer Perceptron (MLP) to arrive at different expression categories encompassing anger, sadness, fear, disgust, surprise and happy. We investigated our proposed system on Cohn-Kanade (CK+), JAFFE, MMI and MUG databases to conduct and validate our experiment and establish its performance superiority over other existing competitors. [ABSTRACT FROM AUTHOR]

The detection of small surface abnormalities on large complex free-form surfaces represents a significant challenge. Often surfaces abnormalities are less than a millimeter square in area but, must be located on surfaces of multiple meters square. To achieve consistent, cost effective and fast inspection, robotic or automated inspection systems are highly desirable. The challenge with automated inspection systems is to create a robust and accurate system that is not adversely affected by environmental variation. Robot-mounted laser line scanner systems can be used to acquire surface measurements, in the form of a point cloud 1 1 PC: point cloud, PCA: principal component analysis, CAD: computer-aided design, ROC: receiving operating characteristic, TP: true positives, FP: false positive, FN: false negative, TPR: true positive rate, TNR: true negative rate, SVM: support vector machine, CV: cross validation. (PC), from large complex geometries. This paper addresses the challenge of how surface abnormalities can be detected based on PC data by considering two different analysis strategies. First, an unsupervised thresholding strategy is considered, and through an experimental study the factors that affect abnormality detection performance are considered. Second, a robust supervised abnormality detection strategy is proposed. The performance of the proposed robust detection algorithm is evaluated experimentally using a realistic test scenario including a complex surface geometry, inconsistent PC quality and variable PC noise. Test results of the unsupervised analysis strategy shows that besides the abnormality size, the laser projection angle and laser lines spacing play an important role on the performance of the unsupervised detection strategy. In addition, a compromise should be made between the threshold value and the sensitivity and specificity of the results. [ABSTRACT FROM AUTHOR]

Background and objective The aim of this study was to assess the utility of bispectrum-based oximetry approaches as a complementary tool to traditional techniques in the screening of pediatric sleep apnea-hypopnea syndrome (SAHS). Methods 298 blood oxygen saturation (SpO 2 ) signals from children ranging 0–13 years of age were recorded during overnight polysomnography (PSG). These recordings were divided into three severity groups according to the PSG-derived apnea hypopnea index (AHI): AHI < 5 events per hour (e/h), 5 ≤ AHI < 10 e/h, AHI ≥ 10 e/h. For each pediatric subject, anthropometric variables, 3% oxygen desaturation index ( ODI 3) and spectral features from power spectral density (PSD) and bispectrum were obtained. Then, the fast correlation-based filter (FCBF) was applied to select a subset of relevant features that may be complementary, excluding those that are redundant. The selected features fed a multiclass multi-layer perceptron (MLP) neural network to build a model to estimate the SAHS severity degrees. Results An optimum subset with features from all the proposed methodological approaches was obtained: variables from bispectrum, as well as PSD, ODI3, Age , and Sex . In the 3-class classification task, the MLP model trained with these features achieved an accuracy of 76.0% and a Cohen's kappa of 0.56 in an independent test set. Additionally, high accuracies were reached using the AHI cutoffs for diagnosis of moderate (AHI = 5 e/h) and severe (AHI = 10 e/h) SAHS: 81.3% and 85.3%, respectively. These results outperformed the diagnostic ability of a MLP model built without using bispectral features. Conclusions Our results suggest that bispectrum provides additional information to anthropometric variables, ODI3 and PSD regarding characterization of changes in the SpO 2 signal caused by respiratory events. Thus, oximetry bispectrum can be a useful tool to provide complementary information for screening of moderate-to-severe pediatric SAHS. [ABSTRACT FROM AUTHOR]

During the last decades, many researches and applications have been conducted on machining features modeling and classification. Although several approaches have been used in the domain of Computer-Aided Process Planning (CAPP), the consideration of freeform shapes has not been well explored at present. This paper aims to provide an up-to-date review of machining features and its conformance with STEP-NC (STEP compliant Numerical Control). A new method based on discrete shape representation, differential geometry and graph theory is developed for freeform machining feature characterization and classification. The effectiveness of the proposed method is demonstrated through an example of a blade part. [ABSTRACT FROM AUTHOR]

In a collaboration environment, it is a challenge how to effectively share the information needed for collaboration while protecting other confidential information in a product assembly model. In this paper, an innovative encryption approach for assembly models to support collaboration is presented. This approach is content based encryption and effective for the secure sharing of feature-based assembly models. In the approach, a classification algorithm for features in an assembly model to be shared or protected during collaboration has been first developed. An encryption algorithm for a feature has been then designed to ensure the parameterization, topological and geometrical validity, and self-adaptability of the encrypted feature. An algorithm for parts with multiple encryption features has been developed. Based on the above algorithms, parts are finally assembled and the geometry and topology of the assembling structure are kept un-changed to enhance collaborators’ interoperability. The characteristics and innovations of the approach include: (1) the approach is feature based, integrative into the main-stream commercial Computer Aided Design (CAD) systems, and flexible to meet various users’ needs for encrypting features selected by users during collaboration, (2) in the approach, the topological and geometrical validity of an assembly model after encryption is maintained to ensure effective collaboration on the assembly, and (3) the approach is parametrically controlled through adjusting position and size parameters so as to ensure the user friendliness of using the approach. A case study with complex geometries and assembly structures has been used to validate the effectiveness and robustness of the approach in industrial applications. [ABSTRACT FROM AUTHOR]

The basic Non-negation Sparse Coding (NNSC) algorithm behaves both advantages of basic Sparse Coding (SC) and Non-negative Matrix Factorization (NMF). It has been proved to be quite more reasonable than SC in explaining the receptive fields of simple cells in V1 in the primary visual system of mammals, and also suitable to process the larger and high dimension data. However, it ignores class information constraint of data, and when it is used in classification task, the advantage is not outstanding. To get better class efficiency, considering the within-class and between class dispersion constraint rule, a new NNSC algorithm based on dispersion constraint denoted by DCB-NNSC one is put forward in this paper, which combines the feature discriminability constraint supervised by classification task and the maximized sparseness criteria in the cost function. At the same time, to ensure the self-adaptive sparse measurement, the Normal Inverse Gaussian Density (NIG) density model is used as sparse penalty function of features. In test, the property of feature extraction is first discussed by implementing the image reconstruction task by using features extracted, and then, the feature recognition capability is further explored by using several classifiers. In the test, test images are provided by the PolyU database. Simulation results show that DCB-NNSC algorithm can capture some significant receptive fields with clearer sparsity and image structure, which are not only efficient in extracting image features, but also favor well the image classification task. Furthermore, comparing DCB-NNSC algorithm discussed here with basic NNSC, experimental results show that DCB-NNSC algorithm is indeed effective and has a good potential in the research task of feature extraction and classification for pattern recognition. [ABSTRACT FROM AUTHOR]

In this study, we evaluated the use of synthetic aperture radar (SAR) and multispectral data to detect aquaculture waterbodies in Southern Bangladesh to quantify fish production on a national scale. For this purpose, we developed an object-based framework comprised of three sequential stages: 1) water detection, 2) feature segmentation, and 3) feature classification. Techniques such as Edge-Otsu for binary thresholding, edge detection with convolution filters, and various supervised and unsupervised machine learning methods were used as part of a workflow. We found that ensemble products combining individual subproducts resulted in higher overall accuracy for water detection (overall detection rate around 60%) and waterbodies classification (overall accuracies up to 79%). Moreover, we showed that SAR data and shape indices played important roles in better-discriminating waterbodies. However, limitations in edge detection outcomes affected the identification of small and isolated aquaculture waterbodies, especially those integrated into rice fields, or in areas with trees. • Information about rapidly growing aquaculture in developing countries is limited. • Synthetic aperture radar and multispectral data were used for detecting aquaculture waterbodies. • Ensemble products have overall high accuracy for water detection/waterbodies classification. • Limitations in edge detection affected the identification of small and isolated waterbodies. [ABSTRACT FROM AUTHOR]

A signal from a pulsar can be decomposed into a set of features. This set is a unique signature for a given pulsar. It can be used to decide whether a pulsar is newly discovered or not. Features can be constructed from coefficients of a wavelet decomposition. Two types of wavelet based pulsar features are proposed. The energy based features reflect the multiscale distribution of the energy of coefficients. The singularity based features first classify the signals into a class with one peak and a class with two peaks by exploring the number of the straight wavelet modulus maxima lines perpendicular to the abscissa, and then implement further classification according to the features of skewness and kurtosis. Experimental results show that the wavelet based features can gain comparatively better performance over the shape parameter based features not only in the clustering and classification, but also in the error rates of the recognition tasks. [ABSTRACT FROM AUTHOR]

Acoustic borehole televiewer (BHTV) logs provide direct observations of lithology, structure and in situ stress in reservoirs, essential for successful well targeting and field management. Analyses of BHTV logs acquired in twenty-three high temperature (≤288 °C) geothermal wells in the Taupo Volcanic Zone, New Zealand, resulted in the modification of BHTV processing techniques and the creation of a descriptive feature classification for hydrothermally altered, volcano-sedimentary-basement type reservoirs lacking other complementary information common in hydrocarbons or lower temperature geothermal systems. Lithological characteristics observed on these BHTV logs are presented, alongside an assessment of the reliability of the structural measurements and image interpretation. [ABSTRACT FROM AUTHOR]

We propose a novel pervasive system to recognise human daily activities from a wearable device. The system is designed in a form of reading glasses, named ‘Smart Glasses’, integrating a 3-axis accelerometer and a first-person view camera. Our aim is to classify subject’s activities of daily living (ADLs) based on their vision and head motion data. This ego-activity recognition system not only allows caretakers to track on a specific person (such as disabled patient or elderly people), but also has the potential to remind/warn people with cognitive impairments of hazardous situations. We present the following contributions: a feature extraction method from accelerometer and video; a classification algorithm integrating both locomotive (body motions) and stationary activities (without or with small motions); a novel multi-scale dynamic graphical model for structured classification over time. In this paper, we collect, train and validate our system on two large datasets: 20 h of elder ADLs datasets and 40 h of patient ADLs datasets, containing 12 and 14 different activities separately. The results show that our method efficiently improves the system performance (F-Measure) over conventional classification approaches by an average of 20%–40% up to 84.45%, with an overall accuracy of 90.04% for elders. Furthermore, we also validate our method on 30 patients with different disabilities, achieving an overall accuracy up to 77.07%. [ABSTRACT FROM AUTHOR]

Abstract: The structure of data is important to the recognition of data. It is a fundamental question how measures and complements the structure of multi-features, because the fusion structure of multi-features is more complete than that of the single feature. To settle the question, we propose three methods for feature structure fusion in feature vectors or feature vector spaces. Firstly, the applicability of the different metric is analyzed. Secondly, optimization questions of various features are constructed based on manifold learning methods. Finally, multiple target optimization questions are transformed to a single target optimization question, and the principle of feature structure fusion is uncovered. In the classification of shape analysis and human action recognition, it is proven that structure fusion methods are effective. [Copyright &y& Elsevier]

Many different algorithms have been proposed for the extraction of features with a range of applications. In this work, we present Tensor-Cuts: a novel framework for feature extraction and classification from images which results in the simultaneous extraction and classification of multiple feature types (surfaces, curves and joints). The proposed framework combines the strengths of tensor encoding, feature extraction using Gabor Jets, global optimization using Graph-Cuts, and is unsupervised and requires no thresholds. We present the application of the proposed framework in the context of road extraction from satellite images, since its characteristics makes it an ideal candidate for use in remote sensing applications where the input data varies widely. We have extensively tested the proposed framework and present the results of its application to road extraction from satellite images. [ABSTRACT FROM AUTHOR]

Abstract: This paper presents classification, representation and extraction of deformation features in sheet-metal parts. The thickness is constant for these shape features and hence these are also referred to as constant thickness features. The deformation feature is represented as a set of faces with a characteristic arrangement among the faces. Deformation of the base-sheet or forming of material creates Bends and Walls with respect to a base-sheet or a reference plane. These are referred to as Basic Deformation Features (BDFs). Compound deformation features having two or more BDFs are defined as characteristic combinations of Bends and Walls and represented as a graph called Basic Deformation Features Graph (BDFG). The graph, therefore, represents a compound deformation feature uniquely. The characteristic arrangement of the faces and type of bends belonging to the feature decide the type and nature of the deformation feature. Algorithms have been developed to extract and identify deformation features from a CAD model of sheet-metal parts. The proposed algorithm does not require folding and unfolding of the part as intermediate steps to recognize deformation features. Representations of typical features are illustrated and results of extracting these deformation features from typical sheet metal parts are presented and discussed. [Copyright &y& Elsevier]

Abstract: Since cable members are the major structural components of cable bridges, they should be properly inspected for surface damage and inside defects such as corrosion and/or breakage of wires. This study introduces an efficient image-based damage detection system that can automatically identify damages to the cable surface through image processing techniques and pattern recognition. The damage detection algorithm combines image enhancement techniques with principal component analysis (PCA) algorithm. Images from three cameras attached to a cable climbing robot are wirelessly transmitted to a server computer located on a stationary cable support. To improve the overall quality of the images, this study utilizes an image enhancement method together with a noise removal technique. Next the input images are projected into PCA sub-space, the Mahalanobis square distance is used to determine the distances between the input images and sample patterns. The smallest distance is found to be a match for an input image. The proposed damage detection algorithm was verified through laboratory tests on three types of cables. Results of the tests showed that the proposed system could be used to detect damage to bridge cables. [Copyright &y& Elsevier]

Abstract: In this paper, we present an image retrieval technique for specific objects based on salient regions. The salient regions we select are invariant to geometric and photometric variations. Those salient regions are detected based on low level features, and need to be classified into different types before they can be applied on further vision tasks. We first classify the selected regions into four types including blobs, edges and lines, textures, and texture boundaries, by using the correlations with the neigbouring regions. Then, some specific region types are chosen for further object retrieval applications. We observe that regions selected from images of the same object are more similar to each other than regions selected from images of different objects. Correlation is used as the similarity measure between regions selected from different images. Two images are considered to contain the same object, if some regions selected from the first image are highly correlated to some regions selected from the second image. Two data sets are employed for experiment: the first data set contains human face images of a number of different people and is used for testing the retrieval algorithm on distinguishing specific objects of the same category; and the second data set contains images of different objects and is used for testing the retrieval algorithm on distinguishing objects of different categories. The results show that our method is very effective on specific object retrieval. [Copyright &y& Elsevier]

The global drive towards a sustainable energy future is giving rise to rapidly increasing penetration of variable renewable energy into modern power grids. This creates a need for the assessment, characterization and classification of renewable energy resources in the context of the operational challenges posed by large-scale grid integration of renewable energy. This investigation explores a methodology for classifying wind energy resources, using feature vectors defined in terms of the statistical properties of the wind resource for Time-of-Use energy demand periods. Results are presented for the geographic areas associated with the South African Renewable Energy Development Zones, using a mesoscale wind atlas dataset as the resource input. The cluster formations obtained with the Time-of-Use feature vector approach are compared with results obtained by clustering the temporal power profiles using the k-means algorithm. It is shown that cluster formations obtained with the respective inputs exhibit distinct differences, especially with reference to the spatial granularity and geographical dispersion of the clusters. It is concluded that the proposed Time-of-Use feature vector approach offers advantages for application as a classification and data partitioning methodology for spatiotemporal wind profiles. • Translated wind power profiles to statistical energy Time-of-Use feature vectors. • Reducion of a high dimensionality temporal wind power dataset. • Statistically reduced dataset retaining critical underlying characteristics. • Comparison of temporospatial inputs with statistical clustering methodology. [ABSTRACT FROM AUTHOR]

• An improved version of the World Cup Optimization Algorithm is introduced. • Chaos theory is utilized for increasing the convergence and the exploration part of the world cup optimization algorithm. • A new configuration based on feature classification is introduced for skin cancer diagnosis. • The proposed optimized method is compared with 4 new methods for showing its superiority. Skin cancer has grown significantly over the past decades, and the importance of its initial treatment is increasing day by day. This study aims to use a computer-aided diagnosis automatic system for the exact diagnosis of cancerous cases. In the present study, after applying pre-processing operations including contrast enhancement, image thresholding, and mathematical morphology for highlighting the important areas, feature extraction based on 20 different features has been applied. Then, for decreasing the system complexity, a new improved version of the world cup optimization algorithm is utilized for the features pruning. After optimal selection of the features, the injected into a support vector machine as a classifier to determine the cancerous areas. Experimental results are compared by different new algorithms including the classic SVM method, hybrid optimized neural networks based on gray wolf optimizer and particle swarm optimization algorithm, and also convolutional neural network. Results obtained for skin lesions images from two different databases indicate that the proposed method has higher efficiency than other published methods, demonstrating even better performance than the well-known classification method, the convolutional neural networks. The proposed method presents the highest correct detection rate - 92.64 for ACS images and 87.5 for ISIC images with very low false acceptance rates (4.41 % and 9 % for ACS and ISIC images, respectively) and even less false rejection rates, this is, 2.94 % on ACS images and 3.5 % on ISIC images. These results prove that the proposed method may be reliably used to diagnose skin cancer through dermoscopy images. [ABSTRACT FROM AUTHOR]

Commercial lightning detection networks offer detailed lightning stroke data upon subscription. Alternatively, the lightning monitoring system can be deployed at lower cost using Very Low Frequency (VLF) signal reception. With that, a dual-class high-intensity lightning recognition algorithm was developed using VLF signal features to identify the severity of lightning occurrence in the vicinity of 500 km from a single VLF signal path. The algorithm combined feature extraction, lightning detection, and lightning severity classification modules to identify two lightning severity classes based on the detected lightning peak current (by Lightning Detection System Network). For each lightning stroke sample, 20 VLF signal features (in terms of wavelet coefficient) in five signal parameters of four different levels of frequency bands were extracted as pattern representation prior to the classification. A combination of bagged and boosted trees ensemble classifiers achieved a lightning detection rate of 64.0% and correct lightning severity classification of 60.3%. These outstanding results exceeded lightning detection rate of between 20% and 40% in related past studies that applied the early VLF events technique. Good design practices such as cross-validation, feature selection, and ensemble classifiers contribute good generalisation and unbiased classification for small dataset. • Power of signal is dominant features in lightning detection. • Zero crossing rate feature useful in lightning recognition system. • Minimum 26.9% lightning detection rate improvement from Early VLF Event technique. • Features optimisation and ensemble classifiers assist small dataset classification. [ABSTRACT FROM AUTHOR]

• In this work, we presented a general feature detection and classification method with single wavelength HSRL at 532 nm, as a detailed reference of the further data analysis for other HSRLs. • To reduce false detection, the threshold profile of the feature detection is chosen to be varied with the signal-to-noise ratio of lidar signal. • Validation experiment is performed with the HSRL system at Zhejiang University. The obtained aerosol optical depth meets well with that from the sun photometer (5.9% in terms of bias error). • Measurements performed over eastern China extend the database of aerosols near the Yangtze River Delta region. The comparison of optical properties between our results and established literatures demonstrates that our measurements and method are reasonably good and satisfactory. A ground-based high-spectral-resolution lidar (HSRL), operated at 532 nm wavelength, has been developed at Zhejiang University (ZJU) for aerosols and clouds studies. This lidar provides vertical profiles of aerosol scattering ratio together with lidar ratio and particle depolarization ratio at 532 nm. These single wavelength HSRL observations can be used to automatically identify the features attributed to dust/polluted dust, urban/smoke, maritime aerosols, as well as ice and liquid water cloud during day and night. Developed algorithm was applied to extended data set obtained during field campaigns performed at coastal city Zhoushan and metropolis Hangzhou over eastern China. To evaluate the practical performance of feature detection, the obtained aerosol optical depth (AOD) from ZJU HSRL has been compared with that from the sun photometers and they agree very well (5.9% in terms of bias error). Also, aerosol classification obtained with presented algorithm agrees well with previous established literatures. Furthermore, based on our method and extensive data sets from ZJU HSRL, measurements performed over eastern China extend the database of aerosols especially with respect to urban and marine and dust near the Yangtze River Delta region. [ABSTRACT FROM AUTHOR]

• The sudden pain, which has not been so common with other pain such as the cold pain and chronic pain, is researched and a complete set of pain experiment platform is built. • Not only the traditional classic methods such as Power Spectral Density (PSD) and Support Vector Machine (SVM), but also the novel SBELM is utilized to get the higher accuracy by optimizing the features. • Every channel has been compared and the preferable channels are selected, which could provide guidance to doctors to treat the illness of patients effectively. Pain is a sensory phenomenon when the body hurts and receptors are stimulated. Although pain activates the body's protective mechanism, some excessive pain reactions will damage the nearby biological tissues, and mostly it will bring people severe mental distress. In particular, some individuals with cognitive impairment, like the infants, are unable to describe their own pain, and thereby the disease will be delayed. These types of pain require intervention and relief and sudden pain belongs to one of them. Therefore, this paper proposes on a method of sudden pain recognition based on resting state EEG signals. This method can recognize the presence of sudden pain, distinguish the location of pain and can be effectively applied to disease diagnosis. The platform of sudden pain stimulation and EEG acquisition system was designed and built, and a series of experiments were carried out for different subjects. We preprocessed the raw EEG signals and extracted features via Power Spectral Density (PSD) and Multifractal Detrended Fluctuation Analysis (MF-DFA). We also utilized the Support Vector Machine (SVM), Sparse Bayesian Extreme Learning Machine (SBELM) and D-S Evidence Theory to do classification, utilizing 10-Fold Cross-validation. The results suggested that the accuracy of judging the presence of pain was up to 89.3 % on average, accuracy of pain location discrimination was up to 81.3 % on average, and accuracy of cross validation was up to 90.1 %. [ABSTRACT FROM AUTHOR]

• An improved version of the Whale Optimization Algorithm is introduced. • Chaos theory is utilized for convergence improvement of WOA. • New configuration for feature classification is introduced. • The method is compared with 4 new methods for showing its superiority. Brain tumor is an abnormal growth of cells in the brain that its diagnosis in the early stages can help us to prevent the dangers of the next stage. In this paper, a new meta-heuristic based methodology is presented for the early diagnosis of the brain tumor to prevent this objection. The proposed method includes three main phases including background removing, feature extraction, and classification based on multilayer perceptron neural network. Here, an improved model of the whale optimization algorithm based on the chaos theory and logistic mapping technique is employed to the optimal selection of the features and the classification stages. The performance analysis of the presented method is compared with some existing methods. Final results showed that based on analyzing CDR, FAR, and FRR as testifying indices, the proposed method has better results than the other similar methods. [ABSTRACT FROM AUTHOR]

High intensity lightning strike can lead to Very Low Frequency (VLF) signal reception. The lack of VLF signal feature characterisation creates the need for commercial lightning dataset subscription, so as to determine the causative lightning strike. This paper presents a signal feature analysis of VLF signal for high intensity lightning severity classification purpose. As such, the analysis was comprised of dual stages, namely feature extraction and feature classification, for two-class recognition. In the feature extraction stage, Discrete Wavelet Transform was implemented to decompose the VLF signal into four frequency bands. In each frequency band, five-signal parameter analysis was executed to yield twenty signal features. Next, the classification stage was executed to determine the best signal feature combination and the most dominant signal feature that generated the highest classification accuracy within the five classifiers. The simulation revealed that the most optimum signal feature combination was achieved via Boosted Trees classifier with classification accuracies of 70% and 50% for training and test datasets, respectively. Class prediction in the skewed test dataset appeared consistent with the test data classification accuracy. The Zero Crossing Rate between 30 and 60 kHz emerged as the most dominant signal feature due to its impact on all classifiers' best signal feature arrangement. The proposed recognition model enables the detection of high intensity lightning classes through the use of VLF signal features. • Zero Crossing Rate is the dominant signal feature to classify lightning severity. • Abrupt perturbation (Energy Entropy) has the least influence on lightning severity. • Boosted Trees is the best classifier in classification accuracy and consistency. • The largest sample to feature ratio does not yield highest classification accuracy. [ABSTRACT FROM AUTHOR]