Misclassified Data Research Articles

Augmenting human-guided progressive learning with machine vision systems for robust surface defect detection

Machine vision systems commonly utilize Convolutional Neural Networks (CNNs) for in-line surface defect detection of manufactured components. The prediction abilities of vision-based inspection systems deteriorate with time as the defect detection model trained on fixed image datasets fails to accommodate deviations. This paper proposes a human-guided progressive learning approach that systematically imparts learning of new features to the CNN-powered vision-based defect detection system. The approach augments the surface defect detection model with human intelligence, using an intuitive user interface to address model drift. The human expert monitors the trained model performance under specific conditions leading to the change of characteristics during implementation, identifies misclassifications, and initiates re-training. The algorithm accumulates misclassified data till a pre-defined threshold level is reached or a human expert terminates inspection. The misclassified results merge with the original datasets for progressive re-training using a strategy similar to the base model development. The present work utilizes pre-trained CNN Efficientnet-b0 to develop the surface defect detection model for tapered roller inspection through transfer learning. It is concluded that the progressive re-training improves defect detection performance and reduces misclassifications. The Matthews Correlation Coefficient (MCC) score, derived from the confusion matrix, showed improvement from 0.6 to 0.82 after four iterations. A cross-model benchmarking study is also performed to show the versatility of the proposed approach. The present work demonstrated that the human-guided progressive learning approach can provide adaptability to vision-based surface defect detection utilizing deep learning algorithms and enhance system performance during real-time implementation.

A fully interpretable stacking fuzzy classifier with stochastic configuration-based learning for high-dimensional data

This study proposes a stacking fuzzy classifier with stochastic configuration-based learning that can achieve higher training and testing performances and sound interpretability of fuzzy rules. By using the understandable first-order Takagi–Sugeno–Kang fuzzy system, we initially stack each successive subclassifier on both the remaining misclassified training data and the corresponding outputs of the previous subclassifier. Subsequently, a Stacking Fuzzy Classifier with Fully Interpretable and Short fuzzy Rules (FISR-SFC) further improves its prediction by linearly aggregating the outputs of all the subclassifiers. FISR-SFC trains each subclassifier using the proposed stochastic configuration-based learning procedure to utilize its training excellence on gradually smaller misclassified training data and simultaneously maintain the full interpretability of each subclassifier. Experimental results on twelve benchmarking datasets reveal that FISR-SFC is at least comparable to and even better than the comparative classifiers in terms of average testing accuracy/G-mean and/or short rules with full interpretability.

Speech Emotion Recognition Incorporating Relative Difficulty and Labeling Reliability.

Emotions in speech are expressed in various ways, and the speech emotion recognition (SER) model may perform poorly on unseen corpora that contain different emotional factors from those expressed in training databases. To construct an SER model robust to unseen corpora, regularization approaches or metric losses have been studied. In this paper, we propose an SER method that incorporates relative difficulty and labeling reliability of each training sample. Inspired by the Proxy-Anchor loss, we propose a novel loss function which gives higher gradients to the samples for which the emotion labels are more difficult to estimate among those in the given minibatch. Since the annotators may label the emotion based on the emotional expression which resides in the conversational context or other modality but is not apparent in the given speech utterance, some of the emotional labels may not be reliable and these unreliable labels may affect the proposed loss function more severely. In this regard, we propose to apply label smoothing for the samples misclassified by a pre-trained SER model. Experimental results showed that the performance of the SER on unseen corpora was improved by adopting the proposed loss function with label smoothing on the misclassified data.

Bagging Algorithm Based on Possibilistic Data Interpolation for Brain-Computer Interface

Recently, brain-computer interfaces (BCIs) and brain-machine interfaces have garnered the attention of researchers. Based on connections with external devices, external computers and machines can be controlled by brain signals measured via near-infrared spectroscopy (NIRS) or electroencephalograph devices. Herein, we propose a novel bagging algorithm that generates interpolation data around misclassified data using a possibilistic function, to be applied to BCIs. In contrast to AdaBoost, which is a conventional ensemble learning method that increases the weight of misclassified data to incorporate them with high probability to the next datasets, we generate interpolation data using a membership function centered on misclassified data and incorporate them into the next datasets simultaneously. The interpolated data are known as virtual data herein. By adding the virtual data to the training data, the volume of the training data becomes sufficient for adjusting the discriminate boundary more accurately. Because the membership function is defined as a possibility distribution, this method is named the bagging algorithm based on the possibility distribution. Herein, we formulate a bagging-type ensemble learning based on the possibility distribution and discuss the usefulness of the proposed method for solving simple calculations using NIRS data.

Experimental Indication to Improve the NN Learning Accuracy by Integrity Constraints From the NN Training Data

Various approaches to improve the classification rate of neural networks (NN) exist. Nevertheless, the application of integrity constraints to mitigate this rate is novel. This paper investigates the effectiveness of integrity constraints (ICs or short: constraints) to improve the performance of neural networks (NNs). This applies in particular to data reduction in training data. The study starts with the application of ICs to the initial NN classification, focusing on the development of data set-specific constraints. These constraints are created by machine learning algorithms, such as multiple linear regression. The method consists of applying these constraints to the misclassified data sets from different tests with the aim of reducing the misclassification rate. The effectiveness of this approach is quantified by comparing the original misclassification rates with those after applying the IC, where a significant reduction was observed in three different test cases. For example, one test case can be described as significant: In Test 1, the misclassification rate decreased from 0.78% to 0.19%, which corresponds to a reduction of 75.6%. Similar improvements were observed in the subsequent tests, underlining the potential of ICs to improve classification accuracy. Thus, this study provides convincing evidence that the NNIC approach is a valuable tool for mitigating misclassification problems in neural network applications. The combination of training a NN and subsequently apply ICs from the training data (NNIC approach) is new and the experimental results indicate evidence for improving the classification rate.

Generative adversarial deep learning in images using Nash equilibrium game theory

A generative adversarial learning (GAL) algorithm is presented to overcome the manipulations that take place in adversarial data and to result in a secured convolutional neural network (CNN). The main objective of the generative algorithm is to make some changes to initial data with positive and negative class labels in testing, hence the CNN results in misclassified data. An adversarial algorithm is used to manipulate the input data that represents the boundaries of learner’s decision-making process. The algorithm generates adversarial modifications to the test dataset using a multiplayer stochastic game approach, without learning how to manipulate the data during training. Then the manipulated data is passed through a CNN for evaluation. The multi-player game consists of an interaction between adversaries which generates manipulations and retrains the model by the learner. The Nash equilibrium game theory (NEGT) is applied to Canadian Institute for Advance Research (CIFAR) dataset. This was done to produce a secure CNN output that is more robust to adversarial data manipulations. The experimental results show that proposed NEGT-GAL achieved a grater mean value of 7.92 and takes less wall clock time of 25,243 sec. Therefore, the proposed NEGT-GAL outperforms the compared existing methods and achieves greater performance.

Driving Pressure: What Is the Harm?

Driving Pressure: What Is the Harm?

A hybrid approach based on mathematical modelling and improved online learning algorithm for data classification

This paper proposes a novel hybrid classifier method using a mixed-integer linear programming (MILP) model and an improved online learning algorithm for general fuzzy min–max neural networks (IOL_GFMM). IOL_GFMM is used to generate initial hyper-boxes for the MILP model in order to improve its efficiency. This paper is the first attempt that integrates the fuzzy min–max neural networks and the MILP model. The applicability of the proposed hybrid model is proven on both real and artificial datasets. Performance evaluation parameters of the novel hybrid classifier are classification accuracy, training time, number of iterations, and number of misclassified data. Results show that the proposed hybrid method provides satisfying improvements in performance evaluation parameters compared with the pure MILP model. Our proposed approach achieves to speed the training time up to 76 times and decreases the number of misclassified data by up to 17 times. Results of artificial datasets show another contribution that the proposed hybrid approach is very efficient in order to classify data samples not only when they belong to separate regions but also when there exists significant overlap.

Post-processing refined ECG delineation based on 1D-UNet

The Electrocardiography (ECG) serves as a standard method for diagnosing cardiovascular disease due to its minimal risk, affordable price and simple application. Clinical information is embedded implicitly in the intervals, amplitude or morphology of different waves that represent the heart-beat cycle in the form of ECG signals. Manually delineating the ECG signals to locate the boundaries of the P, QRS, and T waves in each heart-beat typically requires a long time professional training. To facilitate the delineation and to increase precision, we propose a post-processing refined ECG delineation method that takes full advantage of the morphological information of a heartbeat cycle both in classifying each sample point in the ECG recording using 1D-UNet and determining the boundaries of these waveforms. To proceed, the ECG signals are split into pieces of only one heart beat cycle before sending into the 1D-UNet. These single beat-annotated ECG segments enable the network better extract the local and global features of different waveforms in the ECG, which gives rise to a very precise categorization of each sample point. The post-processing algorithm then uses the morphological information of ECG signal to get rid of the influence of the misclassified data points on the extraction of the onset/offset of different wave components. Tests carried on the two public ECG databases, i.e., the LUDB and QTDB show satisfactory delineation performance, with sensitivities 99.88% and 99.48%, respectively. These results suggest possible applications in wearable and wireless devices for health monitoring.

A Novel Hybrid Ensemble Clustering Technique for Student Performance Prediction

Educational Data Mining (EDM) is a branch of data mining that focuses on extraction of useful knowledge from data generated through academic activities at school, college or at university level. The extracted knowledge can help to perform the academic activities in a better way, so it is useful for students, parents and institutions themselves. One common activity in EDM is students grade prediction with an aim to identify weak or at-risk students. An early identification of such students helps to take supportive measures that may help students to improve. Among a vast number of approaches available in this field, this study mainly focuses on generating a smarter dataset through reduced feature set without compromising the number of records in it and then producing an approach which combines the strengths of classification and clustering for better prediction results. In this study it has been identified that individual features have distinct effect and that removing misclassified data can affect the overall results. Backward selection is adopted using Pearson correlation as a metric to produce smarter dataset with lesser attributes and better accuracy in prediction. After feature set selection, we have applied EMT (Ensemble Meta-Based Tree Model) classification on it to identify best performing classifiers from five families of classifiers. In hybrid approach, first the ensemble clustering is applied on smart dataset and then EMT classification is applied to reevaluate the un-clustered data, which gives a boost in performance and provides us an accuracy of 93%.

Bilevel hyperparameter optimization for support vector classification: theoretical analysis and a solution method

Support vector classification (SVC) is a classical and well-performed learning method for classification problems. A regularization parameter, which significantly affects the classification performance, has to be chosen and this is usually done by the cross-validation procedure. In this paper, we reformulate the hyperparameter selection problem for support vector classification as a bilevel optimization problem in which the upper-level problem minimizes the average number of misclassified data points over all the cross-validation folds, and the lower-level problems are the l_1-loss SVC problems, with each one for each fold in T-fold cross-validation. The resulting bilevel optimization model is then converted to a mathematical program with equilibrium constraints (MPEC). To solve this MPEC, we propose a global relaxation cross-validation algorithm (GR–CV) based on the well-know Sholtes-type global relaxation method (GRM). It is proven to converge to a C-stationary point. Moreover, we prove that the MPEC-tailored version of the Mangasarian–Fromovitz constraint qualification (MFCQ), which is a key property to guarantee the convergence of the GRM, automatically holds at each feasible point of this MPEC. Extensive numerical results verify the efficiency of the proposed approach. In particular, compared with other methods, our algorithm enjoys superior generalization performance over almost all the data sets used in this paper.

Detection of COVID-19 from Deep Breathing Sounds Using Sound Spectrum with Image Augmentation and Deep Learning Techniques

The COVID-19 pandemic is one of the most disruptive outbreaks of the 21st century considering its impacts on our freedoms and social lifestyle. Several methods have been used to monitor and diagnose this virus, which includes the use of RT-PCR test and chest CT/CXR scans. Recent studies have employed various crowdsourced sound data types such as coughing, breathing, sneezing, etc., for the detection of COVID-19. However, the application of artificial intelligence methods and machine learning algorithms on these sound datasets still suffer some limitations such as the poor performance of the test results due to increase of misclassified data, limited datasets resulting in the overfitting of deep learning methods, the high computational cost of some augmentation models, and varying quality feature-extracted images resulting in poor reliability. We propose a simple yet effective deep learning model, called DeepShufNet, for COVID-19 detection. A data augmentation method based on the color transformation and noise addition was used for generating synthetic image datasets from sound data. The efficiencies of the synthetic dataset were evaluated using two feature extraction approaches, namely Mel spectrogram and GFCC. The performance of the proposed DeepShufNet model was evaluated using a deep breathing COSWARA dataset, which shows improved performance with a lower misclassification rate of the minority class. The proposed model achieved an accuracy, precision, recall, specificity, and f-score of 90.1%, 77.1%, 62.7%, 95.98%, and 69.1%, respectively, for positive COVID-19 detection using the Mel COCOA-2 augmented training datasets. The proposed model showed an improved performance compared to some of the state-of-the-art-methods.

Microgrid Fault Detection and Classification Based on the Boosting Ensemble Method With the Hilbert-Huang Transform

In this paper, a sequential ensemble of intelligence-based methods is used for fault detection and classification in microgrids. The proposed scheme is recommended because of the impracticality of conventional fault detection and protection due to microgrid dynamic behavior and dependency of traditional methods on faults current level or impedance. These methods use the collective decision of learners to increase their accuracy. This is done by distributing the knowledge among the classifiers. The proposed ensemble method is called Brownboost. Its main advantage over its counterparts is that it uses a nonconvex optimization method. This makes it robust to overfitting and applicable and practical for a real-world noisy or misclassified data. In addition, a signal processing method called the Hilbert-Huang transform was chosen for feature extraction from signals transient behavior to reduce the noise sensitivity. In this method, current difference of the both ends of line is selected to decompose various types of adaptive basis signal processing method. The results are validated in IEC test microgrid with various noise penetration level and synchronization delays. Compared to traditional strong classifiers, this method is easy to tune and program and is robust to overfitting. Moreover, it can be justified for real-world imperfect data.

Misclassification Analysis of Elementary School Accreditation Data in Ambon City Using Multivariate Adaptive Regression Spline

Many classification methods have been developed, one of which is the Multivariate Adaptive Regression Spline (MARS) method. MARS is one of the classification methods in the form of a combination of Recursive Partitioning Regression (RPR) and the spline method that is able to process high-dimensional and large-sized data and process data with continuous or binary response variables. The purpose of this study was to measure the misclassification of elementary school accreditation in Ambon city using the MARS method. This study uses accreditation data with the results of eight components of accreditation in elementary schools that have accreditation A (group 1) and accreditation B (group 2) in Ambon city. To evaluate the classification method used the APER classification error measure. The best classification result from the MARS method is when using a combination of BF=32, MI=3, MO=1 because it produces a minimum Generalized Cross Validation (GCV) of 0.066 and information is obtained that the correct classification data is 181 and the misclassified data is 10. Based on the results of the analysis, the size of the APER classification error is 5.23%, which can be said that the MARS method is good or statistically significant for classifying elementary schools in Ambon City based on their accreditation rating.

183 Advancing Structural and Systematic Equity in Epidemiological Analyses of Large Datasets

OBJECTIVES/GOALS: Health inequities represent complex structural and systematic processes that lead to disparate outcomes for populations or subgroups within populations. This project aims to improve the available structural and systematic approaches to the study of such inequities at the population level. METHODS/STUDY POPULATION: Using examples from diabetes research, two critical factors that may impact the validity or utility of health equity models will be examined; and proposed methodological approaches to offsetting potentially resulting biases will be offered. The factors include: (1) inequitably missing and misclassified data in large datasets and (2) the presumed positioning of socially constructed variables such as race, ethnicity, and gender within modeled structural and systematic mechanisms. This examination intersects theories and praxis in epidemiological modeling and health equity promotion with the goal of advancing rigorous, equity-focused epidemiological methods. RESULTS/ANTICIPATED RESULTS: Inequitably missing and misclassified data are generally expected to obscure inequities. Treatment of missing or misclassified data as informative measures of inequity is expected to partially offset this bias. The implicitly modeled components of socially constructed variables are expected to be non-uniform across structural and systematic mechanisms of inequity. Models that apply these variables as informatively heterogeneous constructs, using multi-phase analyses to test modeling assumptions and to assess intersectionality, may provide better context about the mechanisms by which inequity has been distributed and, perhaps, by which equity may be achieved. DISCUSSION/SIGNIFICANCE: Equitable epidemiological methods are essential to the advancement of evidence-based health equity on the population level. Potential structural or systematic inequities in large-scale datasets and traditional data analyses should be thoughtfully reviewed through a health equity lens.

The Mutually Reinforcing Cycle Of Poor Data Quality And Racialized Stereotypes That Shapes Asian American Health.

The Asian American health narrative reflects a long history of structural racism in the US and the complex interplay of racialized history, immigrant patterns, and policies regarding Asians in the US. Yet owing to systematic issues in data collection including missing or misclassified data for Asian Americans and practices that lead to indiscriminate grouping of unlike individuals (for example, Chinese, Vietnamese, and Bangladeshi) together in data systems and pervasive stereotypes of Asian Americans, the drivers and experiences of health disparities experienced by these diverse groups remain unclear. The perpetual exclusion and misrepresentation of Asian American experiences in health research is exacerbated by three racialized stereotypes-the model minority, healthy immigrant effect, and perpetual foreigner-that fuel scientific and societal perceptions that Asian Americans do not experience health disparities. This codifies racist biases against the Asian American population in a mutually reinforcing cycle. In this article we describe the poor-quality data infrastructure and biases on the part of researchers and public health professionals, and we highlight examples from the health disparities literature. We provide recommendations on how to implement systems-level change and educational reform to infuse racial equity in future policy and practice for Asian American communities.

Early Detection of Network Intrusions Using a GAN-Based One-Class Classifier

Early detection of network intrusions is a very important factor in network security. However, most studies of network intrusion detection systems utilize features for full sessions, making it difficult to detect intrusions before a session ends. To solve this problem, the proposed method uses packet data for features to determine if packets are malicious traffic. Such an approach inevitably increases the probability of falsely detecting normal packets as an intrusion or an intrusion as normal traffic for the initial session. As a solution, the proposed method learns the patterns of packets that are unhelpful in order to classify network intrusions and benign sessions. To this end, a new training dataset for Generative Adversarial Network (GAN) is created using misclassified data from an original training dataset by the LSTM-DNN model trained using the original one. The GAN trained with this dataset has ability to determine whether the currently received packet can be accurately classified in the LSTM-DNN. If the GAN determines that the packet cannot be classified correctly, the detection process is canceled and will be tried again when the next packet is received. Meticulously designed classification algorithm based on LSTM-DNN and validation model using GAN enable the proposed algorithm to accurately perform network intrusion detection in real time without session termination or delay time for collecting a certain number of packets. Various experiments confirm that the proposed method can detect intrusions very early (before the end of the session) while maintaining detection performance at a level similar to that of the existing methods.

A simulation-extrapolation approach for regression analysisof misclassified current status data with the additive hazards model.

Current status data arise when each subject is observed only once and the failure time of interest is only known to be either smaller or larger than the observation time rather than observed exactly. For the situation, due to the use of imperfect diagnostic tests, the failure status could often suffer misclassification or one observes misclassified data, which may result in severely biased estimation if not taken into account. In this article, we discuss regression analysis of such misclassified current status data arising from the additive hazards model, and a simulation-extrapolation (SIMEX) approach is developed for the estimation. Furthermore, the asymptotic properties of the proposed estimators are established, and a simulation study is conducted to assess the empirical performance of the method, which indicates that the proposed procedure performs well. In particular, it can correct the estimation bias given by the naive method that ignores the existence of misclassification. An application to a medical study on gonorrhea is also provided.

DETEKSI DINI KEBAKARAN HUTAN DAN LAHAN MEMANFAATKAN EKSTRAKSI EXIF PADA INFORMASI GAMBAR BERBASIS PENGOLAHAN CITRA

Forest fire detection system is one of important tools in preventing and mitigating forest and land fires. In Indonesia, the detection of forest and land fires relies on hotspot information captured from satellites. However, the location obtained by the satellite has a horizontal error of 2 km from the ground check data. Therefore, these information are less relevant to the actual location. In this research, an android app is proposed to extract Exchangeable Image Format (EXIF) photo metadata. The metadata has image information such as latitude and longitude, to obtain the location of forest fires reported by the application user. In addition, this research implemented one of the image processing methods to classify fire and smoke in images of fires. Color filtering method is used based on the color space of Red Green Blue (RGB), Hue Saturation Value (HSV) and YCbCr. This classification process aims to ease the burden on the admin in confirming user reports. The results of the fire and smoke classification process are described using a confusion matrix. This matrix produces an accuracy rate of 75%, a precision of 80% and a recall of 80% for a fire classification and an accuracy of 70%, a precision of 92% and a recall of 87% for smoke classification. There are 25% and 30% of misclassified data of fire and smoke. This is because the color filtering method classifies each color pixel from the image, therefore many pixels that are not classified as fire or smoke images are classified because there are other objects that have a range of colors to classify fire and smoke

Cryptocurrency Scams: Analysis and Perspectives

Since the inception of Bitcoin in 2009, the market of cryptocurrencies has grown beyond the initial expectations, as witnessed by the thousands of tokenised assets available on the market, whose daily trades exceed dozens of USD billions. The pseudonymity features of cryptocurrencies have attracted the attention of cybercriminals, who exploit them to carry out potentially untraceable scams. The wide range of cryptocurrency-based scams observed over the last ten years has fostered the study on their effects, and the development of techniques to counter them. The research in this field is hampered by various factors. First, there exist only a few public data sources about cryptocurrency scams, and they often contain incomplete or misclassified data. Further, there is no standard taxonomy of scams, which leads to ambiguous and incoherent interpretations of their nature. Indeed, the unavailability of reliable datasets makes it difficult to train effective automatic classifiers that can detect and analyse scams. In this paper, we perform an extensive review of the scientific literature on cryptocurrency scams, which we systematise according to a novel taxonomy. By collecting and homogenising data from different public sources, we build a uniform dataset of thousands of cryptocurrency scams. We build upon this dataset to implement a tool that automatically recognises scams and classifies them according to our taxonomy. We assess the effectiveness of our tool through standard performance metrics. We then analyse the results of the classification, providing key insights about the distribution of scam types, and the correlation between different types. Finally, we propose a set of guidelines that policymakers could follow to improve user protection against cryptocurrency scams.