Individual Class Accuracies Research Articles

Mapping invasive strawberry guava (Psidium cattleianum) in tropical forests of Mauritius with Sentinel-2 and machine learning

ABSTRACT Invasive flora alien species (IAS) pose a serious threat to biodiversity in tropical forests. This study evaluated different classification approaches for remote sensing of the target IAS, strawberry guava (Psidium cattleianum), using open-access satellite imagery to aid research and understanding of IAS impacts in tropical forests. This is likely the first study using freely available open-access imagery to map strawberry guava, and the first mapping of invasive species in Mauritius using remote sensing. Ground-based observations in Black River Gorges National Park (BRGNP), Mauritius, resulted in 4670 ‘ground truth’ samples across three land cover classes representing different levels of strawberry guava canopy cover; 70% of these ground truth data were used for training and model tuning, and 30% were held-out for testing. Classification was performed using Sentinel-2 MSI images and Google Earth Engine. Computation of Jeffries-Matusita distance supported the addition of Grey Level Co-Occurrence Matrix texture measures, and spectral indices (ARI1, and ReNDVI) for classification, since these features increased separability of strawberry guava cover classes substantially over using spectral bands alone. Higher than 80% overall and individual class accuracies were achieved with both Random Forest (RF) and Support Vector Machine (SVM) classification of strawberry guava in BRGNP. RF is recommended for similar future applications for its higher overall accuracy (97.60% ± .20% at 95% confidence), lower variation, and prediction of more homogenous class regions. This study found that strawberry guava canopy cover in BRGNP was highest in central regions and correlated to steeper slopes, with little overall change from 2016 to 2020.

Performance on, and correlates of, the Brief Visuospatial Memory Test—Revised after traumatic brain injury

ABSTRACT Objective This study investigated the performance on, and correlates of, the Brief Visuospatial Memory Test – Revised (BVMT–R) in patients with traumatic brain injury (TBI). Methods Participants included 100 patients with TBI and 100 demographically matched controls. We first used regression analysis to determine predictors of BVMT–R performance in the clinical group. We then used analysis of variance as well as logistic regression to determine how BVMT–R findings differed between the clinical and control groups. Results Injury severity and visuospatial ability both contributed to the prediction of BVMT–R Total Recall and Delayed Recall scores in the TBI group. Mean differences between the TBI and control groups on these variables were statistically significant, but overall individual classification accuracy was limited at 59%. Conclusions The BVMT–R has some clinical utility in the evaluation of patients with TBI but should not be used in isolation.

Effect of Using Different Amounts of Multi-Temporal Data on the Accuracy: A Case of Land Cover Mapping of Parts of Africa Using FengYun-3C Data

Regional or continental-scale land cover mapping requires various amounts of months of multi-temporal satellite data to pick phenological variation in vegetation, enhancing differentiability among surface cover types and improving accuracy. However, little has been addressed about the number of months/multi-temporal images needed to obtain the best result and the impact of using different amounts of these data on the accuracy of individual classes. This work aimed to analyze these effects by utilizing the various amounts of months of time series FengYun-3C (FY-3C) data within one year for land cover mapping of parts of Africa using a random forest classifier. The study area covers roughly one-third of Africa, including eastern, central, and northern parts of the continent. One-year FY-3C ten-day composite images consisting of eleven-band each with 1-km spatial resolution were divided into seven input datasets that comprise stacked images of 1-month, 3-month, 6-month, consecutive 9-month, 12-month, selected images from 12 months using band/feature importance, and selected 9-month. Comparisons of these datasets on independent test samples revealed that overall accuracy, kappa coefficient, and the accuracy of the individual classes generally increase significantly with increasing the number of data/months. However, the highest accuracy and kappa coefficient, 0.86 and 0.83, were obtained by processing selected 9-month imageries. The second maximum accuracy and kappa (0.85 and 0.82,) were found by manipulating 12-month scenes which are the same as the results obtained by applying feature reduction. Although 4% and 5% higher accuracy were achieved by manipulating 3-month and 6-month data relative to 1-month imageries, no variation of accuracy was observed between six- and nine-months of consecutive data, both yielding equal accuracy and kappa value (0.84 and 0.81) indicating redundancy of information. Overall, the high accuracy results show the feasibility of FY-3C data for land cover mapping of Africa.

LSB steganography detection in monochromatic still images using artificial neural networks

Embedding graphic content in multimedia through steganography is a useful and fast practice to hide information. However, detecting the use of this technique is complex and sometimes unsuccessful because variations are not visually perceptible. This article proposes the use of a binary classification model based on artificial neural networks to detect the presence of LSB steganography on monochromatic still images of 256x256 and 8 bits, based on the Standford Genome Project. The steganograms were generated by varying the payload from 0.1 to 0.5 to obtain image pairs of carriers and steganograms. For each steganogram, the following features were extracted from image histograms: kurtosis, skewness, standard deviation, range, median, harmonic mean, Hjorth mobility, and complexity. The results show that the classifier reaches a 91.45% accuracy in detecting LSB steganography when learning from all payloads, as well as a 96.78% individual classification accuracy in the best case with a payload of 0.5.

Forest Mapping Using Classification of Sentinel-2A Imagery for Forest Fire Danger Prediction: a Case Study

Timely and accurate effective forest cover mapping is a prerequisite for predicting forest fire danger. Remote sensing data have an undoubted advantage in mapping the forest cover of territories. The paper compares six trained classification algorithms in order to select the best Sentinel 2A image for a typical forestry in the Baikal region. The conducted comparative analysis has included comparison of the classification accuracy by parametric and nonparametric methods with the default parameters set in the ENVI software. The training sample (Samples data) has been created based on forest management materials. The overall accuracy and the Cohen's kappa coefficient have been used to assess general performance of each algorithm. The accuracy of mapping individual vegetation classes has been assessed using the accuracy of the producer and the user and their combination of F-score. The results of the study can be used when choosing a method for classifying forest vegetation in the Baikal zone and other similar areas by satellite imagery in order to predict forest fire danger.

Injury severity prediction of traffic crashes with ensemble machine learning techniques: a comparative study

A better understanding of injury severity risk factors is fundamental to improving crash prediction and effective implementation of appropriate mitigation strategies. Traditional statistical models widely used in this regard have predefined correlation and intrinsic assumptions, which, if flouted, may yield biased predictions. The present study investigates the possibility of using the eXtreme Gradient Boosting (XGBoost) model compared with few traditional machine learning algorithms (logistic regression, random forest, and decision tree) for crash injury severity analysis. The data used in this study was obtained from the traffic safety department, ministry of transport (MOT) at Riyadh, KSA, and contains 13,546 motor vehicle collisions along 15 rural highways reported between January 2017 to December 2019. Empirical results obtained using k-fold (k = 10) for various performance metrics showed that the XGBoost technique outperformed other models in terms of the collective predictive performance as well as injury severity individual class accuracies. XGBoost feature importance analysis indicated that collision type, weather status, road surface conditions, on-site damage type, lighting conditions, and vehicle type are the few sensitive variables in predicting the crash injury severity outcome. Finally, a comparative analysis of XGBoost based on different performance statistics showed that our model outperformed most previous studies.

AUTOMATIC ARRHYTHMIA DETECTION BASED ON THE ANALYSIS OF ELECTROCARDIOGRAMS WITH DEEP LEARNING

According to the World Health Organization, cardiovascular diseases (CVD) are one of the most common causes of death in the world. The most effective clinical method for visualizing the cardiac electrical activity is electrocardiography (ECG). Automated ECG analysis has been of great interest in the medical researches. The problem of automated detection of cardiac arrhythmias may be reduced to the ECG signals classification. To solve this task such methods were used as Hidden Markov Models (HMM), discrete wavelet transforms (DWT), support vector machine (SVM) etc. Now days, the deep learning models began to play the major role in solving this problem. In this paper, for the classification of ECG signals, a number of models of deep neural networks, including deep convolutional, recurrent based on short-term long memory have been developed and implemented. To improve the classification accuracy of individual classes of the studied data, the CNN-LSTM deep model was built, which combines convolutional and recurrent networks. In addition the following machine learning algorithms were used for ECG signals classification: support vector machine (SVM), decision trees (DT), random forest (RF) and extreme gradient boosting (XGB). To test the performance of the proposed models, MIT-BIH database was used, a freely available dataset that is widely used to evaluate the effectiveness of ECG signal classification algorithms. The results of a comparative analysis of various algorithms for the quality of classification for individual classes showed that machine learning algorithms classify classes with a large volume of samples well. For example, SVM and DT classify samples from class N and Q with an accuracy of 92 and 97%, respectively, while samples from classes S and F are classified with much worse accuracy of 63%. At the same time, analyzing and comparing the performance of various neural network models based on the obtained estimates of the classification accuracy, it can be argued that CNN LSTM model allows not only a high classification accuracy of 99.37%, but also high values of other indicators of classification quality, such as F1- metric, precision, and recall.The proposed algorithms for the automated detection of cardiac arrhythmias can be applied in biomedical applications that analyze the electrocardiogram and help physicians diagnose cardiac arrhythmias more accurately.

A comparative study of machine learning classifiers for injury severity prediction of crashes involving three-wheeled motorized rickshaw

Motorcycles and motorcyclists have a variety of attributes that have been found to be a potential contributor to the high liability of vulnerable road users (VRUs). Vulnerable Road Users (VRUs) that include pedestrians, bicyclists, cycle-rickshaw occupants, and motorcyclists constitute by far the highest share of road traffic accidents in developing countries. Motorized three-wheeled Rickshaws (3W-MR) is a popular public transport mode in almost all Pakistani cities and is used primarily for short trips to carry passengers and small-scale goods movement. Despite being an important mode of public transport in the developing world, little work has been done to understand the factors affecting the injury severity of three-wheeled motorized vehicles. Crash injury severity prediction is a promising research target in traffic safety. Traditional statistical models have underlying assumptions and predefined associations, which can yield misleading results if flouted. Machine learning(ML) is an emerging non-parametric method that can effectively capture the non-linear effects of both continuous and discrete variables without prior assumptions and achieve better prediction accuracy. This research analyzed injury severity of three-wheeled motorized rickshaws (3W-MR) using various machine learning-based identification algorithms, i.e., Decision jungle (DJ), Random Forest (RF), and Decision Tree (DT). Three years of crash data (from 2017 to 2019) was collected from Provincial Emergency Response Service RESCUE 1122 for Rawalpindi city, Pakistan. A total of 2,743 3W-MR crashes were reported during the study period that resulted in 258 fatalities. The predictive performance of proposed ML models was assessed using several evaluation metrics such as overall accuracy, macro-average precision, macro-average recall, and geometric means of individual class accuracies. Results revealed that DJ with an overall accuracy of 83.7 % outperformed the DT and RF-based on a stratified 10-fold cross-validation approach. Finally, Spearman correlation analysis showed that factors such as the lighting condition, crashes involving young drivers (aged 20–30 years), facilities with high-speed limits (over 60 mph), weekday, off-peak, and shiny weather conditions were more likely to worsen injury severity of 3W-MR crashes. The outcomes of this study could provide necessary and essential guidance to road safety agencies, particularly in the study area, for proactive implementation of appropriate countermeasures to curb road safety issues pertaining to three-wheeled motorized vehicles.

Mean response latency indices on the Victoria Symptom Validity Test do not contribute meaningful predictive value over accuracy scores for detecting invalid performance

The utility of the Victoria Symptom Validity Test (VSVT) as a performance validity test (PVT) has been primarily established using response accuracy scores. However, the degree to which response latency may contribute to accurate classification of performance invalidity over and above accuracy scores remains understudied. Therefore, this study investigated whether combining VSVT accuracy and response latency scores would increase predictive utility beyond use of accuracy scores alone. Data from a mixed clinical sample of 163 patients, who were administered the VSVT as part of a larger neuropsychological battery, were analyzed. At least four independent criterion PVTs were used to establish validity groups (121 valid/42 invalid). Logistic regression models examining each difficulty level revealed that all VSVT measures were useful in classifying validity groups, both independently and when combined. Individual predictor classification accuracy ranged from 77.9 to 81.6%, indicating acceptable to excellent discriminability across the validity indices. The results of this study support the value of both accuracy and latency scores on the VSVT to identify performance invalidity, although the accuracy scores had superior classification statistics compared to response latency, and mean latency indices provided no unique benefit for classification accuracy beyond dimensional accuracy scores alone.

Classification of evoked responses to inverted faces reveals both spatial and temporal cortical response abnormalities in Autism spectrum disorder

The neurophysiology of face processing has been studied extensively in the context of social impairments associated with autism spectrum disorder (ASD), but the existing studies have concentrated mainly on univariate analyses of responses to upright faces, and, less frequently, inverted faces. The small number of existing studies on neurophysiological responses to inverted face in ASD have used univariate approaches, with divergent results. Here, we used a data-driven, classification-based, multivariate machine learning decoding approach to investigate the temporal and spatial properties of the neurophysiological evoked response for upright and inverted faces, relative to the neurophysiological evoked response for houses, a neutral stimulus. 21 (2 females) ASD and 29 (4 females) TD participants ages 7 to 19 took part in this study. Group level classification accuracies were obtained for each condition, using first the temporal domain of the evoked responses, and then the spatial distribution of the evoked responses on the cortical surface, each separately. We found that classification of responses to inverted neutral faces vs. houses was less accurate in ASD compared to TD, in both the temporal and spatial domains. In contrast, there were no group differences in the classification of evoked responses to upright neutral faces relative to houses. Using the classification in the temporal domain, lower decoding accuracies in ASD were found around 120 ms and 170 ms, corresponding the known components of the evoked responses to faces. Using the classification in the spatial domain, lower decoding accuracies in ASD were found in the right superior marginal gyrus (SMG), intra-parietal sulcus (IPS) and posterior superior temporal sulcus (pSTS), but not in core face processing areas. Importantly, individual classification accuracies from both the temporal and spatial classifiers correlated with ASD severity, confirming the relevance of the results to the ASD phenotype.

Automated facial recognition for wildlife that lack unique markings: A deep learning approach for brown bears.

Emerging technologies support a new era of applied wildlife research, generating data on scales from individuals to populations. Computer vision methods can process large datasets generated through image‐based techniques by automating the detection and identification of species and individuals. With the exception of primates, however, there are no objective visual methods of individual identification for species that lack unique and consistent body markings. We apply deep learning approaches of facial recognition using object detection, landmark detection, a similarity comparison network, and an support vector machine‐based classifier to identify individuals in a representative species, the brown bear Ursus arctos. Our open‐source application, BearID, detects a bear’s face in an image, rotates and extracts the face, creates an “embedding” for the face, and uses the embedding to classify the individual. We trained and tested the application using labeled images of 132 known individuals collected from British Columbia, Canada, and Alaska, USA. Based on 4,674 images, with an 80/20% split for training and testing, respectively, we achieved a facial detection (ability to find a face) average precision of 0.98 and an individual classification (ability to identify the individual) accuracy of 83.9%. BearID and its annotated source code provide a replicable methodology for applying deep learning methods of facial recognition applicable to many other species that lack distinguishing markings. Further analyses of performance should focus on the influence of certain parameters on recognition accuracy, such as age and body size. Combining BearID with camera trapping could facilitate fine‐scale behavioral research such as individual spatiotemporal activity patterns, and a cost‐effective method of population monitoring through mark–recapture studies, with implications for species and landscape conservation and management. Applications to practical conservation include identifying problem individuals in human–wildlife conflicts, and evaluating the intrapopulation variation in efficacy of conservation strategies, such as wildlife crossings.

Instruction Sequence Identification and Disassembly Using Power Supply Side-Channel Analysis

Embedded systems are prone to leak information via side-channels associated with their physical internal activity, such as power consumption, timing, and faults. Leaked information can be analyzed to extract sensitive data and devices should be assessed for such vulnerabilities. Side-channel power-supply leakage from embedded devices can also provide information regarding instruction-level activity for control code executed on these devices. Methods proposed to disassemble instruction-level activity via side-channel leakage have not addressed issues related to pipelined multi-clock-cycle architectures, nor have proven robustness or reliability. The problem of detecting malicious code modifications while not obstructing the sequence of instructions being executed needs to be addressed. In this article, instruction sequences being executed on a general-purpose pipelined computing platform are identified and instructions that make up these sequences are classified based on hardware utilization. Individual instruction classification results using a fine-grained classifier is also presented. A dynamic programming algorithm was applied to detect the boundaries of instructions in a sequence with a 100 percent accuracy. A unique aspect of this technique is the use of multiple power supply pin measurements to increase precision and accuracy. To demonstrate the robustness of this technique, power leakage data from ten target FPGAs programmed with a prototype of the pipelined architecture was analyzed and classification accuracies averaging 99 percent were achieved with instructions labeled based on hardware utilization. Individual instruction classification accuracies above 90 percent were achieved using a fine-grained classifier. Classification accuracies were also verified when a target FPGA was subjected to different controlled temperatures. The classification accuracies on discrete (ASIC) pipelined-architecture microcontrollers was 97 percent.

Performance of different machine learning algorithms on satellite image classification in rural and urban setup

In many countries, the increasing population drives the spatiotemporal land use and land cover change (LULCC) at a higher rate, causing heterogeneous landscape. Hence, the LULCC is a dynamic and frequent process causing fragmented land cover, and therefore, extensive research on LULCC pattern is necessary at different spatial and temporal scales. Moreover, it is essential to identify appropriate algorithms to detect LULCC in such fragmented areas. Furthermore, the rate of change is different in rural and urban areas. Hence, the main goal of the study was to describe the performance of different machine learning algorithms on three different spatial and multispectral satellite image classification in rural and urban extents. We carried out atmospheric and geometric correction. To achieve this, we acquired and processed a set of moderate and finer resolution images (Landsat-8, Sentinel-2, and Planet images) having similar phenological stages to finding out the suitable algorithms that provide better performance in LULC classification. Random forest, Support Vector Machine (SVM), and their combined strength (stacked algorithms) were applied on Landsat-8, Sentinel-2, and Planet images separately to assess individual and overall class accuracy of the images. Two unique sets of training data were generated to classify the 6 (3 sensors x 2 locations) imagery. We find that the Sentinel-2 image performs best among the three images. Among the three different algorithms, SVM showed comparatively better results. In both Bhola and Dhaka, the SVM on Sentinel image performed best with an overall accuracy of 0.969, 0.983, and overall kappa of 0.948, and 0.968, respectively. This study illustrates the role of algorithms on LULCC studies for a better understanding of the changes that occurred in the rural and urban areas. These findings assist planners, remote sensing scientists and decision-makers to choose a suitable image classification algorithm in monitoring rapidly changing, fragmented, and diverse landscape in Bangladesh and elsewhere in the world.

Deep feature extraction and classification of breast ultrasound images

Controlled despeckling (structure/edges/feature preservation with smoothing the homogeneous areas) is a desired pre-processing step for the design of computer-aided diagnostic (CAD) systems using ultrasound images as the presence of speckle noise masks diagnostically important information making interpretation difficult even for experienced radiologist. For efficiently classifying the breast tumors, the conventional CAD system designs use hand-crafted features. However, these features are not robust to the variations in size, shape and orientation of the tumors resulting in lower sensitivity. Thus deep feature extraction and classification of breast ultrasound images have recently gained attention from research community. The deep networks come with an advantage of directly learning the representative features from the images. However, these networks are difficult to train from scratch if the representative training data is small in size. Therefore transfer learning approach for deep feature extraction and classification of medical images has been widely used. In the present work the performance of four pre-trained convolutional neural networks VGG-19, SqueezeNet, ResNet-18 and GoogLeNet has been evaluated for differentiating between benign and malignant tumor types. From the results of the experiments, it is noted that CAD system design using GoogLeNet architecture for deep feature extraction followed by correlation based feature selection and fuzzy feature selection using ANFC-LH yields highest accuracy of 98.0% with individual class accuracy value of 100% and 96% for benign and malignant classes respectively. For differentiating between the breast tumors, the proposed CAD system design can be utilized in routine clinical environment.

Decision Tree Algorithms for Developing Rulesets for Object-Based Land Cover Classification

Decision tree (DT) algorithms are important non-parametric tools used for land cover classification. While different DTs have been applied to Landsat land cover classification, their individual classification accuracies and performance have not been compared, especially on their effectiveness to produce accurate thresholds for developing rulesets for object-based land cover classification. Here, the focus was on comparing the performance of five DT algorithms: Tree, C5.0, Rpart, Ipred, and Party. These DT algorithms were used to classify ten land cover classes using Landsat 8 images on the Copperbelt Province of Zambia. Classification was done using object-based image analysis (OBIA) through the development of rulesets with thresholds defined by the DTs. The performance of the DT algorithms was assessed based on: (1) DT accuracy through cross-validation; (2) land cover classification accuracy of thematic maps; and (3) other structure properties such as the sizes of the tree diagrams and variable selection abilities. The results indicate that only the rulesets developed from DT algorithms with simple structures and a minimum number of variables produced high land cover classification accuracies (overall accuracy > 88%). Thus, algorithms such as Tree and Rpart produced higher classification results as compared to C5.0 and Party DT algorithms, which involve many variables in classification. This high accuracy has been attributed to the ability to minimize overfitting and the capacity to handle noise in the data during training by the Tree and Rpart DTs. The study produced new insights on the formal selection of DT algorithms for OBIA ruleset development. Therefore, the Tree and Rpart algorithms could be used for developing rulesets because they produce high land cover classification accuracies and have simple structures. As an avenue of future studies, the performance of DT algorithms can be compared with contemporary machine-learning classifiers (e.g., Random Forest and Support Vector Machine).

Design of Feedforward Neural Networks in the Classification of Hyperspectral Imagery Using Superstructural Optimization

Artificial Neural Networks (ANNs) have been used in a wide range of applications for complex datasets with their flexible mathematical architecture. The flexibility is favored by the introduction of a higher number of connections and variables, in general. However, over-parameterization of the ANN equations and the existence of redundant input variables usually result in poor test performance. This paper proposes a superstructure-based mixed-integer nonlinear programming method for optimal structural design including neuron number selection, pruning, and input selection for multilayer perceptron (MLP) ANNs. In addition, this method uses statistical measures such as the parameter covariance matrix in order to increase the test performance while permitting reduced training performance. The suggested approach was implemented on two public hyperspectral datasets (with 10% and 50% sampling ratios), namely Indian Pines and Pavia University, for the classification problem. The test results revealed promising performances compared to the standard fully connected neural networks in terms of the estimated overall and individual class accuracies. With the application of the proposed superstructural optimization, fully connected networks were pruned by over 60% in terms of the total number of connections, resulting in an increase of 4% for the 10% sampling ratio and a 1% decrease for the 50% sampling ratio. Moreover, over 20% of the spectral bands in the Indian Pines data and 30% in the Pavia University data were found statistically insignificant, and they were thus removed from the MLP networks. As a result, the proposed method was found effective in optimizing the architectural design with high generalization capabilities, particularly for fewer numbers of samples. The analysis of the eliminated spectral bands revealed that the proposed algorithm mostly removed the bands adjacent to the pre-eliminated noisy bands and highly correlated bands carrying similar information.

ML Algorithm Performance to Classify MCS Schemes During UACN Link Adaptation

This research classifies the modulation and coding rate for link adaptation in Underwater Acoustic Communications Networks (UACNs). Recently, the UACN has become a promising technology for military, commercial, and civilian applications, as well as scientific research. However, we should minimize the dataset dimension for real-time implementation due to the sensor nodes’ energy limitations in the underwater environment. We used an Incheon sea trial’s measured dataset of 18 features, applying Principal Component Analysis (PCA) to select the dominant eigenvalue components in order to reduce the curse of dimensionality, and then selected 11 parameters. After that, we applied Machine Learning (ML) algorithms with different combinations of the parameters to separately classify the modulation and the coding rate and measured both individual and overall classification accuracy. The findings are compared with two Taean sea trial datasets with 11 features to finalize the selected parameters for link adaptation. For modulation classification, we observed 96.83% accuracy with the K-nearest Neighbors (KNN) algorithm in three-parameter and two-parameter cases. In coding rate classification, we found 100% accuracy with the KNN algorithm using the same three-parameter case. However, for the best fit among the three datasets, we finalized another three parameters at the expense of accuracy. To find the optimum threshold values for all modulation and coding rate labels, we used Rule-based (RB) 2D and 3D analysis. However, with a hard limit on non-overlapping data, at best, 35.51% classification accuracy was found for a 1/3 coding rate (Turbo code) with QPSK modulation, which showed much less reliability for RB analysis in a UACN, so it is not useful in this regard. Besides, our analysis shows data independence in the Doppler Spread (DS) and the Frequency Shift (FS) , mitigating the time-variability channel’s challenge. We use the Gaussian distribution plot, a confusion matrix, multi-dimensional scatter plots, interpolated plots to analyze the data.

Consecutive spatial-spectral framework for remote sensing image classification

This paper contributes an artificial intelligence classifier for hyperspectral images by simultaneously employing spectral and spatial information in kernel functions and using the stochastic relaxation of spectral and spatial energy. The spectral and spatial features are integrated with the Consecutive Spectral-Spatial/Spatial-Spectral (CS2) kernel which exhibits great flexibility in the combination of spectral and spatial mapping functions. The overall classification accuracy and individual class accuracies of the proposed classier were compared with some well-known spectral-spatial kernels in two empirical hyperspectral benchmark datasets. The experimental findings demonstrated the flexibility and effectivity of CS2 kernel in the field of simultaneous spectral-spatial processing. The computational complexity of the proposed CS2 kernel-based Support Vector Machine (CS3VM) is adequacy equal to the required amount of execution time of published approaches in spectral-spatial kernel literature. The advantage of the CS2 kernel is its generalization capability in spectral-spatial kernel machines. An additional post-processing step is exploited to preserve the edge of regions and to generate homogenized classification maps based on probabilistic weighted spectral-spatial energy minimization. Different types of evaluation metrics were used to prove the accuracy of the proposed framework under different training conditions and scenarios. The CS2 kernel machines produced very competitive results and the proposed probabilistic stochastic relaxation approach improved the average accuracy, overall accuracy and the kappa statistic with Indian Pines and University of Pavia hyperspectral datasets. The novel framework provides smooth enough homogeneous regions classification maps with sharp edges on the boundaries and meaningfully enhances the classification performance with limited training samples.

A comprehensive decision support approach for credit scoring

Purpose The purpose of this paper is to provide a comprehensive decision support approach in credit risk assessment. Design/methodology/approach A comprehensive decision support approach is proposed for credit scoring and prediction. The predictive performance of the new approach has been investigated by using data including number and text. Findings The results demonstrate that the proposed approach achieves better and more stable classification accuracy than the single classifiers in most cases. Meanwhile, the prediction accuracy of individual classifiers is also improved by the proposed approach. Originality/value This study provides a comprehensive model for credit risk scoring and provides valuable information to the existing literature on credit scoring by using artificial intelligence.

End-to-End Ensemble Learning by Exploiting the Correlation Between Individuals and Weights.

Ensemble learning performs better than a single classifier in most tasks due to the diversity among multiple classifiers. However, the enhancement of the diversity is at the expense of reducing the accuracies of individual classifiers in general and, thus, how to balance the diversity and accuracies is crucial for improving the ensemble performance. In this paper, we propose a new ensemble method which exploits the correlation between individual classifiers and their corresponding weights by constructing a joint optimization model to achieve the tradeoff between the diversity and the accuracy. Specifically, the proposed framework can be modeled as a shallow network and efficiently trained by the end-to-end manner. In the proposed ensemble method, not only can a high total classification performance be achieved by the weighted classifiers but also the individual classifier can be updated based on the error of the optimized weighted classifiers ensemble. Furthermore, the sparsity constraint is imposed on the weight to enforce that partial individual classifiers are selected for final classification. Finally, the experimental results on the UCI datasets demonstrate that the proposed method effectively improves the performance of classification compared with relevant existing ensemble methods.