Multi-class Classification Approaches Research Articles

Differential Diagnosis of Diabetic Foot Osteomyelitis and Charcot Neuropathic Osteoarthropathy with Deep Learning Methods.

Our study aims to evaluate the potential of a deep learning (DL) algorithm for differentiating the signal intensity of bone marrow between osteomyelitis (OM), Charcot neuropathic osteoarthropathy (CNO), and trauma (TR). The local ethics committee approved this retrospective study. From 148 patients, segmentation resulted in 679 labeled regions for T1-weighted images (comprising 151 CNO, 257 OM, and 271 TR) and 714 labeled regions for T2-weighted images (consisting of 160 CNO, 272 OM, and 282 TR). We employed both multi-class classification (MCC) and binary-class classification (BCC) approaches to compare the classification outcomes of CNO, TR, and OM. The ResNet-50 and the EfficientNet-b0 accuracy values were computed at 96.2% and 97.1%, respectively, for T1-weighted images. Additionally, accuracy values for ResNet-50 and the EfficientNet-b0 were determined at 95.6% and 96.8%, respectively, for T2-weighted images. Also, according to BCC for CNO, OM, and TR, the sensitivity of ResNet-50 is 91.1%, 92.4%, and 96.6% and the sensitivity of EfficientNet-b0 is 93.2%, 97.6%, and 98.1% for T1, respectively. For CNO, OM, and TR, the sensitivity of ResNet-50 is 94.9%, 83.6%, and 97.9% and the sensitivity of EfficientNet-b0 is 95.6%, 85.2%, and 98.6% for T2, respectively. The specificity values of ResNet-50 for CNO, OM, and TR in T1-weighted images are 98.1%, 97.9%, and 94.7% and 98.6%, 97.5%, and 96.7% in T2-weighted images respectively. Similarly, for EfficientNet-b0, the specificity values are 98.9%, 98.7%, and 98.4% and 99.1%, 98.5%, and 98.7% for T1-weighted and T2-weighted images respectively. In the diabetic foot, deep learning methods serve as a non-invasive tool to differentiate CNO, OM, and TR with high accuracy.

AI and semantic ontology for personalized activity eCoaching in healthy lifestyle recommendations: a meta-heuristic approach

BackgroundAutomated coaches (eCoach) can help people lead a healthy lifestyle (e.g., reduction of sedentary bouts) with continuous health status monitoring and personalized recommendation generation with artificial intelligence (AI). Semantic ontology can play a crucial role in knowledge representation, data integration, and information retrieval.MethodsThis study proposes a semantic ontology model to annotate the AI predictions, forecasting outcomes, and personal preferences to conceptualize a personalized recommendation generation model with a hybrid approach. This study considers a mixed activity projection method that takes individual activity insights from the univariate time-series prediction and ensemble multi-class classification approaches. We have introduced a way to improve the prediction result with a residual error minimization (REM) technique and make it meaningful in recommendation presentation with a Naïve-based interval prediction approach. We have integrated the activity prediction results in an ontology for semantic interpretation. A SPARQL query protocol and RDF Query Language (SPARQL) have generated personalized recommendations in an understandable format. Moreover, we have evaluated the performance of the time-series prediction and classification models against standard metrics on both imbalanced and balanced public PMData and private MOX2-5 activity datasets. We have used Adaptive Synthetic (ADASYN) to generate synthetic data from the minority classes to avoid bias. The activity datasets were collected from healthy adults (n = 16 for public datasets; n = 15 for private datasets). The standard ensemble algorithms have been used to investigate the possibility of classifying daily physical activity levels into the following activity classes: sedentary (0), low active (1), active (2), highly active (3), and rigorous active (4). The daily step count, low physical activity (LPA), medium physical activity (MPA), and vigorous physical activity (VPA) serve as input for the classification models. Subsequently, we re-verify the classifiers on the private MOX2-5 dataset. The performance of the ontology has been assessed with reasoning and SPARQL query execution time. Additionally, we have verified our ontology for effective recommendation generation.ResultsWe have tested several standard AI algorithms and selected the best-performing model with optimized configuration for our use case by empirical testing. We have found that the autoregression model with the REM method outperforms the autoregression model without the REM method for both datasets. Gradient Boost (GB) classifier outperforms other classifiers with a mean accuracy score of 98.00%, and 99.00% for imbalanced PMData and MOX2-5 datasets, respectively, and 98.30%, and 99.80% for balanced PMData and MOX2-5 datasets, respectively. Hermit reasoner performs better than other ontology reasoners under defined settings. Our proposed algorithm shows a direction to combine the AI prediction forecasting results in an ontology to generate personalized activity recommendations in eCoaching.ConclusionThe proposed method combining step-prediction, activity-level classification techniques, and personal preference information with semantic rules is an asset for generating personalized recommendations.

Assessing the relevance of mental health factors in fibromyalgia severity: A data-driven case study using explainable AI

Background and objectiveFibromyalgia is a chronic disease that causes pain and affects patients' quality of life. Current treatments focus on pharmacological therapies for pain reduction. However, patients' psychological well-being is also affected, with depression and pain catastrophizing being common. This research addresses the clinicians’ need to assess the influence of mental health factors on FM severity compared to pain factors. MethodsA co-development study between FM clinicians and data scientists analyzed data from 166 FM-diagnosed patients to assess the influence of mental health factors on FM severity in comparison to pain factors. The study used the Polysymptomatic Distress Scale (PDS) and Fibromyalgia Impact Questionnaire (FIQ) as FM severity indicators and collected 15 variables including regarding demographics, pain intensity perceived, and mental health factors. The team used an author's developed framework to identify the optimal FM severity classifier and explainability by selecting a number of features that lead to obtaining the best classification result. Machine learning classifiers employed in the framework were: decision trees, logistic regression, support vector machines, random forests, AdaBoost, extra trees, and RUSBoost. Explainability analyses were conducted using the following explainable AI techniques: SHapley Additive exPlanations (SHAP), Partial Dependence Plot (PDP), and Mean Decrease Impurity (MDI). ResultsA balanced random forest with 6 features achieved the best performance with PDS (AUC_ROC, mean = 0.81, std = 0.07). Being FIQ the target variable, due to the imbalance in FM severity levels, a binary and a multiclass classification approaches were considered achieving the optimal performance, respectively, a logistic regression classifier (AUC_ROC, mean = 0.83, std = 0.08) with 6 selected features, and a random forest (AUC_ROC, mean = 0.91, std = 0.04) with 8 selected features. Next, the explainability analysis determined mental health factors were found to be more relevant than pain perceived factors for FM severity. ConclusionsThis study’s findings, validated by clinicians, are potentially aligned with FM international guidelines that promote non-pharmacological interventions such as promoting mental well-being of FM patients.

An ensemble-based faults detection and diagnosis approach for determining faults severities at whole-building level

Data-driven fault detection and diagnosis (FDD) approaches rely on operational data even when in-depth knowledge of the studied system is lacking. Simulated data is affordable and can cover numerous fault types and severities. Developing FDD models with a single fault severity has limitations on the model's generalisability owing to evolution of faults with severities. Thus, machine learning models must be individually trained on diverse scenarios. Current studies on building FDD are often at system levels, thereby leaving significant knowledge gaps when there is interest in energy performance of an entire building. This research presents a data-driven FDD approach for classifying several building faults using different ensemble multi-class classification approaches. Additionally, the impact of noise on training datasets is investigated. The XGBoost classifier achieved the highest classification accuracy for the considered faults during validation and testing stages, with low sensitivity to noise, thereby demonstrating a promising potential for wider scale deployment.

Product Review Classification using Machine Learning and Statistical Data Analysis

The aim of the paper is to implement and analyze the machine learning models for product review dataset. The project focuses on binary classification, multi-class classification, and clustering approaches to analyze and categorize product reviews. The performance of the models over each of the five classification tasks is measured by the 5-fold cross-validation scores over the training data.

A whole-building data-driven fault detection and diagnosis approach for public buildings in hot climate regions

Fault detection and diagnosis (FDD) approaches comprise three main pillars: model-based, knowledge-based, and data-driven strategies. Data-driven approaches prioritise operational data and do not necessitate in-depth understanding of the system's background; yet, significant amounts of data is required, which often poses challenges to researchers. Since simulated data is inexpensive and can run numerous faults types with varying severities and time periods, it has been used in data-driven FDD analysis. However, the majority of FDD approaches are implemented at the system level of buildings. However, most buildings have numerous systems with distinct features. Furthermore, using individualised system-level analysis makes it difficult to see system-to-system relationships. Currently, there is a significant underrepresentation of research that investigate the applications of FDD models under whole-building scenarios, so as to identify a wider range of energy consumption related faults in buildings. Furthermore, since data-driven approaches significantly depend on the quantities of training data, it becomes challenging to diagnose faults that have limited features. As a result, this study diagnoses numerous building systems faults, including single and simultaneous faults with limited features. This is implemented within the context of the whole-building energy performance of religious buildings in hot climatic areas, employing data-driven FDD methodologies. Various multi-class classification approaches were investigated to classify both the normal condition and faulty classes. Furthermore, feature extraction methodologies were compared to quantify their potential for improving the diagnosis. In addition to the classification evaluation metrics, one-way ANOVA and Tukey-Kramer tests were implemented to examine the significance of the reported performance differences. RF classifier obtained highest classification accuracy during validation and testing with about 90%, indicating a promising performance in whole-building faults analysis. The adoption of feature extraction techniques did not improve classification performance, thereby emphasising that some classifiers may perform better with high-dimensional datasets.

Social media data analysis framework for disaster response

This paper presents a social media data analysis framework applied to multiple datasets. The method developed uses machine learning classifiers, where filtering binary classifiers based on deep bidirectional neural networks are trained on benchmark datasets of disaster responses for earthquakes and floods and extreme flood events. The classifiers consist of learning from discrete handcrafted features and fine-tuning approaches using deep bidirectional Transformer neural networks on these disaster response datasets. With the development of the multiclass classification approach, we compare the state-of-the-art results in one of the benchmark datasets containing the largest number of disaster-related categories. The multiclass classification approaches developed in this research with support vector machines provide a precision of 0.83 and 0.79 compared to Bernoulli naïve Bayes, which are 0.59 and 0.76, and multinomial naïve Bayes, which are 0.79 and 0.91, respectively. The binary classification methods based on the MDRM dataset show a higher precision with deep learning methods (DistilBERT) than BoW and TF-IDF, while in the case of UnifiedCEHMET dataset show a high performance for accuracy with the deep learning method in terms of severity, with a precision of 0.92 compared to BoW and TF-IDF method which has a precision of 0.68 and 0.70, respectively.

Incident duration prediction using a bi-level machine learning framework with outlier removal and intra–extra joint optimisation

Predicting the duration of traffic incidents is a challenging task due to the stochastic nature of events. The ability to accurately predict how long accidents will last can provide significant benefits to both end-users in their route choice and traffic operation managers in handling of non-recurrent traffic congestion. This paper presents a novel bi-level machine learning framework enhanced with outlier removal and intra–extra joint optimisation for predicting the incident duration on three heterogeneous data sets collected for both arterial roads and motorways from Sydney, Australia and San-Francisco, U.S.A. Firstly, we use incident data logs to develop a binary classification prediction approach, which allows us to classify traffic incidents as short-term or long-term. We find the optimal threshold between short-term versus long-term traffic incident duration, targeting both class balance and prediction performance while also comparing the binary versus multi-class classification approaches using quantiled duration groups and varying threshold split. Secondly, for more granularity of the incident duration prediction to the minute level, we propose a new intra–extra Joint Optimisation algorithm (IEO-ML) which extends multiple baseline ML models tested against several regression scenarios across the data sets. Final results indicate that: (a) 40–45 min is the best split threshold for identifying short versus long-term incidents and that these incidents should be modelled separately, (b) our proposed IEO-ML approach significantly outperforms baseline ML models in 66% of all cases showcasing its great potential for accurate incident duration prediction. Lastly, we evaluate the feature importance and show that time, location, incident type, incident reporting source and weather at among the top 10 critical factors which influence how long incidents will last.

Identifying large-area mangrove distribution based on remote sensing: A binary classification approach considering subclasses of non-mangroves

Mangroves have tremendous ecological value but are vulnerable to anthropogenic factors and sea level rise. Classification based on remote sensing is the first step for monitoring mangrove distribution in large areas but suffers from misclassifications due to the complexity of mangroves, such as having abundant species within the mangrove class, a large latitudinal span from tropical to subtropical, and mixing with nearby forests at the fringe. Binary classification approaches (i.e., mangroves and non-mangroves) have the potential to adjust the decision surface by adding subclasses similar to the mangrove class into the non-mangrove class in comparison with one-class classification approaches (i.e., separating mangroves from the background), and it demands fewer samples than multi-class classification (i.e., a mangrove class and multiple non-mangrove classes). However, the subclasses of binary classifications have rarely been considered to reduce the sample size with comparable performance to multi-class classification approaches. This paper first tested whether a conversion from either of two existing multi-class classifications to a binary classification could achieve comparable performance to corresponding multi-class classification for identifying mangroves in China in 2018 using the Google Earth Engine platform. The results showed that the conversion from multi-class classifications to binary classification achieved an overall accuracy value of 74.1% and 83.8%, respectively, the second of which was higher than that of the corresponding multi-class classification result with a reduction of sample size by a rate of 72.9%. Such an improved performance is due to the adjustment of subclasses during conversion when evaluated by the same validation dataset. Thus, this study provided a binary classification approach to identifying large-area mangrove distribution with a reduced sample size and comparable performance. This paper further discussed the impact of the feature selection procedure and that of subclass settings to reveal the margins of the proposed approach, the distribution and causes of false positives in binary classification results, and adaptations suitable for further generalizations to case areas outside of China. This study fills a knowledge gap and serves for accurate and consistent large-area mangrove identification based on remote sensing.

ReportAGE: Automatically extracting the exact age of Twitter users based on self-reports in tweets.

Advancing the utility of social media data for research applications requires methods for automatically detecting demographic information about social media study populations, including users’ age. The objective of this study was to develop and evaluate a method that automatically identifies the exact age of users based on self-reports in their tweets. Our end-to-end automatic natural language processing (NLP) pipeline, ReportAGE, includes query patterns to retrieve tweets that potentially mention an age, a classifier to distinguish retrieved tweets that self-report the user’s exact age (“age” tweets) and those that do not (“no age” tweets), and rule-based extraction to identify the age. To develop and evaluate ReportAGE, we manually annotated 11,000 tweets that matched the query patterns. Based on 1000 tweets that were annotated by all five annotators, inter-annotator agreement (Fleiss’ kappa) was 0.80 for distinguishing “age” and “no age” tweets, and 0.95 for identifying the exact age among the “age” tweets on which the annotators agreed. A deep neural network classifier, based on a RoBERTa-Large pretrained transformer model, achieved the highest F1-score of 0.914 (precision = 0.905, recall = 0.942) for the “age” class. When the age extraction was evaluated using the classifier’s predictions, it achieved an F1-score of 0.855 (precision = 0.805, recall = 0.914) for the “age” class. When it was evaluated directly on the held-out test set, it achieved an F1-score of 0.931 (precision = 0.873, recall = 0.998) for the “age” class. We deployed ReportAGE on a collection of more than 1.2 billion tweets, posted by 245,927 users, and predicted ages for 132,637 (54%) of them. Scaling the detection of exact age to this large number of users can advance the utility of social media data for research applications that do not align with the predefined age groupings of extant binary or multi-class classification approaches.

Early Detection of Pancreatic Cancers Using Liquid Biopsies and Hierarchical Decision Structure.

Objective: Pancreatic cancer (PC) is a silent killer, because its detection is difficult and to date no effective treatment has been developed. In the US, the current 5-year survival rate of 11%. Therefore, PC has to be detected as early as possible. Methods and procedures: In this work, we have combined the use of ultrasensitive nanobiosensors for protease/arginase detection with information fusion based hierarchical decision structure to detect PC at the localized stage by means of a simple Liquid Biopsy. The problem of early-stage detection of pancreatic cancer is modelled as a multi-class classification problem. We propose a Hard Hierarchical Decision Structure (HDS) along with appropriate feature engineering steps to improve the performance of conventional multi-class classification approaches. Further, a Soft Hierarchical Decision Structure (SDS) is developed to additionally provide confidences of predicted labels in the form of class probability values. These frameworks overcome the limitations of existing research studies that employ simple biostatistical tools and do not effectively exploit the information provided by ultrasensitive protease/arginase analyses. Results: The experimental results demonstrate that an overall mean classification accuracy of around 92% is obtained using the proposed approach, as opposed to 75% with conventional multi-class classification approaches. This illustrates that the proposed HDS framework outperforms traditional classification techniques for early-stage PC detection. Conclusion: Although this study is only based on 31 pancreatic cancer patients and a healthy control group of 48 human subjects, it has enabled combining Liquid Biopsies and Machine Learning methodologies to reach the goal of earliest PC detection. The provision of both decision labels (via HDS) as well as class probabilities (via SDS) helps clinicians identify instances where statistical model-based predictions lack confidence. This further aids in determining if more tests are required for better diagnosis. Such a strategy makes the output of our decision model more interpretable and can assist with the diagnostic procedure. Clinical impact: With further validation, the proposed framework can be employed as a decision support tool for the clinicians to help in detection of pancreatic cancer at early stages.

Ensemble Method of Convolutional Neural Networks with Directed Acyclic Graph Using Dermoscopic Images: Melanoma Detection Application.

The early detection of melanoma is the most efficient way to reduce its mortality rate. Dermatologists achieve this task with the help of dermoscopy, a non-invasive tool allowing the visualization of patterns of skin lesions. Computer-aided diagnosis (CAD) systems developed on dermoscopic images are needed to assist dermatologists. These systems rely mainly on multiclass classification approaches. However, the multiclass classification of skin lesions by an automated system remains a challenging task. Decomposing a multiclass problem into a binary problem can reduce the complexity of the initial problem and increase the overall performance. This paper proposes a CAD system to classify dermoscopic images into three diagnosis classes: melanoma, nevi, and seborrheic keratosis. We introduce a novel ensemble scheme of convolutional neural networks (CNNs), inspired by decomposition and ensemble methods, to improve the performance of the CAD system. Unlike conventional ensemble methods, we use a directed acyclic graph to aggregate binary CNNs for the melanoma detection task. On the ISIC 2018 public dataset, our method achieves the best balanced accuracy (76.6%) among multiclass CNNs, an ensemble of multiclass CNNs with classical aggregation methods, and other related works. Our results reveal that the directed acyclic graph is a meaningful approach to develop a reliable and robust automated diagnosis system for the multiclass classification of dermoscopic images.

Discrimination of Fresh Tobacco Leaves with Different Maturity Levels by Near-Infrared (NIR) Spectroscopy and Deep Learning.

The maturity affects the yield, quality, and economic value of tobacco leaves. Leaf maturity level discrimination is an important step in manual harvesting. However, the maturity judgment of fresh tobacco leaves by grower visual evaluation is subjective, which may lead to quality loss and low prices. Therefore, an objective and reliable discriminant technique for tobacco leaf maturity level based on near-infrared (NIR) spectroscopy combined with a deep learning approach of convolutional neural networks (CNNs) is proposed in this study. To assess the performance of the proposed maturity discriminant model, four conventional multiclass classification approaches—K-nearest neighbor (KNN), backpropagation neural network (BPNN), support vector machine (SVM), and extreme learning machine (ELM)—were employed for a comparative analysis of three categories (upper, middle, and lower position) of tobacco leaves. Experimental results showed that the CNN discriminant models were able to precisely classify the maturity level of tobacco leaves for the above three data sets with accuracies of 96.18%, 95.2%, and 97.31%, respectively. Moreover, the CNN models with strong feature extraction and learning ability were superior to the KNN, BPNN, SVM, and ELM models. Thus, NIR spectroscopy combined with CNN is a promising alternative to overcome the limitations of sensory assessment for tobacco leaf maturity level recognition. The development of a maturity-distinguishing model can provide an accurate, reliable, and scientific auxiliary means for tobacco leaf harvesting.

Classification of imbalanced hyperspectral images using SMOTE-based deep learning methods

Hyperspectral imaging (HSI) is one of the most advanced methods of digital imaging. This technique differs from RGB images with its wide range of the electromagnetic spectrum. Imbalanced data sets are frequently encountered in machine learning. As a result, the classifier performance may be poor. To avoid this problem, the data set must be balanced. The main motivation in this study is to reveal the difference and effects on the classifier performance between the original imbalanced dataset and the data set modified by balancing methods. In the proposed method, hyperspectral image classification study carried out on Xuzhou Hyspex dataset includes nine-classes including bareland-1, bareland-2, crops-1, crops-2, lake, coals, cement, trees, house-roofs of elements, by using the convolutional neural networks (CNN) and dataset balancing methods comprising the Smote, Adasyn, K-Means, and Cluster. This dataset has been taken from IEEE-Dataport Machine Learning Repository. To classify the hyperspectral image, the convolutional neural networks having different multiclass classification approaches like One-vs-All, One-vs-One. Dataset was splitted in two different ways: %50–%50 Hold-out and 5-Fold Cross-validation. In order to evaluate the performance of the proposed models, the confusion matrix, classification accuracy, precision, recall, and F-Measure have been used. Without the dataset balancing, the obtained classification accuracies are 93.63%, 92.33%, 88.36% for %50–%50 train-test split, and 94.46%, 94%, 92.24% for 5-Fold cross-validation using multi-class classification, One-vs-All, and One-vs-One respectively. After Smote balancing, the obtained classification accuracies are 96.41%, 95.6%, 92.53% for %50–%50 train-test split and 96.49%, 95.64%, 93.38% for 5-Fold cross-validation using multi-class classification, One-vs-All and One-vs-One respectively. After Adasyn balancing, the obtained classification accuracies are 95.86%, 93.62%, 87.05% for %50–%50 train-test split and 96.38%, 95.09%, 91.55% for 5-Fold cross-validation using multi-class classification, One-vs-All and One-vs-One respectively. After K-Means balancing, the obtained classification accuracies are 95.23%, 93.36%, 90.6% for %50–%50 train-test split and 95.74%, 94.72%, 91.94% for 5-Fold cross-validation using multi-class classification, One-vs-All and One-vs-One respectively. After Cluster balancing, the obtained classification accuracies are 94.83%, 94.1%, 90.07% for %50–%50 train-test split and 96.28%, 95.88%, 92.5% for 5-Fold cross-validation using multi-class classification, One-vs-All and One-vs-One respectively. The obtained results have shown that the best model is Smote Balanced 5-CV multiclass classification.

Assessing the influence of point-of-interest features on the classification of place categories

Points of interest (POIs) digitally represent real-world amenities as point locations. POI categories (e.g. restaurant, hotel, museum etc.) play a prominent role in several location-based applications such as social media, navigation, recommender systems, geographic information retrieval tools, and travel-related services. The majority of user queries in these applications center around POI categories. For instance, people often search for the closest pub or the best value-for-money hotel in an area. To provide valid answers to such queries, accurate and consistent information on POI categories is an essential requirement. Nevertheless, category-based annotations of POIs are often missing. The task of annotating unlabeled POIs in terms of their categories — known as POI classification — is commonly achieved by means of machine learning (ML) models, often referred to as classifiers. Central to this task is the extraction of known features from pre-labeled POIs in order to train the classifiers and, then, use the trained models to categorize unlabeled POIs. However, the set of features used in this process can heavily influence the classification results. Research on defining the influence of different features on the categorization of POIs is currently lacking. This paper contributes a study of feature importance for the classification of unlabeled POIs into categories. We define five feature sets that address operation based, review-based, topic-based, neighborhood-based, and visual attributes of POIs. Contrary to existing studies that predominantly use multi-class classification approaches, and in order to assess and rank the influence of POI features on the categorization task, we propose both a multi-class and a binary classification approach. These, respectively, predict the place category among a specified set of POI categories, or indicate whether a POI belongs to a certain category. Using POI data from Amsterdam and Athens to implement and evaluate our study approach, we show that operation based features, such as opening or visiting hours throughout the day, are the most important place category predictors. Moreover, we demonstrate that the use of feature combinations, as opposed to the use of individual features, improves the classification performance by an average of 15%, in terms of F1-score.

Combining frequency and time-domain EEG features for classification of self-paced reach-and-grasp actions.

Brain-computer interfaces (BCIs) might provide an intuitive way for severely motor impaired persons to operate assistive devices to perform daily life activities. Recent studies have shown that complex hand movements, such as reach-and-grasp tasks, can be decoded from the low frequency of the electroencephalogram (EEG). In this work we investigated whether additional features extracted from the frequency-domain of alpha and beta bands could improve classification performance of rest vs. palmar vs. lateral grasp. We analysed two multi-class classification approaches, the first using features from the low frequency time-domain, and the second in which we combined the time-domain with frequency-domain features from alpha and beta bands. We measured EEG of ten participants without motor disability which performed self-paced reach-and-grasp actions on objects of daily life. For the time-domain classification approach, participants reached an average peak accuracy of 65%. For the combined approach, an average peak accuracy of 75% was reached. In both approaches and for all subjects, performance was significantly higher than chance level (38.1%, 3-class scenario). By computing the confusion matrices as well as feature rankings through the Fisher score, we show that movement vs. rest classification performance increased considerably in the combined approach and was the main responsible for the multi-class higher performance. These findings could help the development of BCIs in real-life scenarios, where decreasing false movement detections could drastically increase the end-user acceptance and usability of BCIs.

Evaluating the Performance of Multi-Class and Single-Class Classification Approaches for Mountain Agriculture Extraction Using Time-Series NDVI

Supervised multi-class classification (MCC) approach is widely being used for regional-level land use–land cover (LULC) mapping and monitoring. However, it becomes inefficient if the end user wants to map only one particular class. Therefore, an improved single-class classification (SCC) approach is required for quick and reliable map production purpose. In this regard, the current study attempts to evaluate the performance of MCC and SCC approaches for extracting mountain agriculture area using time-series normalized differential vegetation index (NDVI). At first, samples of eight LULC classes were acquired using Google Earth image, and corresponding temporal signatures (TS) were extracted from time-series NDVI to perform classification using minimum distance to mean (MDM) and spectral angle mapper (i.e., multi-class SAM—MCSAM) under MCC approach. Secondly, under SCC approach, the TS of three agriculture classes (i.e., agriculture, mixed agriculture and plantation) were utilized as a reference to extract agriculture extent using Euclidean distance (ED) and SAM (i.e., single-class SAM—SCSAM) algorithms. The area of all four maps (i.e., MDM—19.77% of total geographical area (TGA), MCSAM—21.07% of TGA, ED—15.23% of TGA, SCSAM—13.85% of TGA) was compared with reference agriculture area (14.54% of TGA) of global land cover product, and SCC-based maps were found to have close agreement. Also, the class-wise detection accuracy was evaluated using random sample point-based error matrix which reveals the better performance of ED-based map than rest three maps in terms of overall accuracy and kappa coefficient.

Deep Age Estimation: From Classification to Ranking

Human age is considered an important biometric trait for human identification or search. Recent research shows that the aging features deeply learned from large-scale data lead to significant performance improvement on facial image-based age estimation. However, age-related ordinal information is totally ignored in these approaches. In this paper, we propose a novel convolutional neural network (CNN)-based framework, ranking-CNN, for age estimation. Ranking-CNN contains a set of basic CNNs, each of which is trained with ordinal age labels. Then, their binary outputs are aggregated for the final age prediction. From a theoretical perspective, we obtain an approximation for the final ranking error, show that it is controlled by the maximum error produced among subranking problems, and thus find a new error bound, which provides helpful guidance for the training and analysis of deep rankers. Based on the new error bound, we theoretically give an explicit formula for the learning of ranking-CNN and demonstrate its convergence using the stochastic approximation method. Moreover, we rigorously prove that ranking-CNN, by considering ordinal relation between ages, is more likely to get smaller estimation errors when compared with multiclass classification approaches. Through extensive experiments, we show that ranking-CNN outperforms other state-of-the-art feature extractors and age estimators on benchmark datasets.

MULTI-LABEL CLASSIFICATION FOR PHYSICAL ACTIVITY RECOGNITION FROM VARIOUS ACCELEROMETER SENSOR POSITIONS

In recent years, the use of accelerometers embedded in smartphones for Human Activity Recognition (HAR) has been well considered. Nevertheless, the role of the sensor placement is yet to be explored and needs to be further investigated. In this study, we investigated the role of sensor placements for recognizing various types of physical activities using the accelerometer sensor embedded in the smartphone. In fact, most of the reported work in HAR utilized traditional multi-class classification approaches to determine the types of activities. Hence, this study was to recognize the activity based on the best sensor placements that are appropriate to the activity performed. The traditional multi-class classification approach required more manual work and was time consuming to run the experiment separately. Thus, this study proposed the multi- label classification technique with the Label Combination (LC) approach in order to tackle this issue. The result was compared with several state-of-the-art traditional multi-class classification approaches. The multi-label classification result significantly outperformed the traditional multi-class classification methods as well as minimized the model build time.

A Novel Wearable Device for Food Intake and Physical Activity Recognition.

Presence of speech and motion artifacts has been shown to impact the performance of wearable sensor systems used for automatic detection of food intake. This work presents a novel wearable device which can detect food intake even when the user is physically active and/or talking. The device consists of a piezoelectric strain sensor placed on the temporalis muscle, an accelerometer, and a data acquisition module connected to the temple of eyeglasses. Data from 10 participants was collected while they performed activities including quiet sitting, talking, eating while sitting, eating while walking, and walking. Piezoelectric strain sensor and accelerometer signals were divided into non-overlapping epochs of 3 s; four features were computed for each signal. To differentiate between eating and not eating, as well as between sedentary postures and physical activity, two multiclass classification approaches are presented. The first approach used a single classifier with sensor fusion and the second approach used two-stage classification. The best results were achieved when two separate linear support vector machine (SVM) classifiers were trained for food intake and activity detection, and their results were combined using a decision tree (two-stage classification) to determine the final class. This approach resulted in an average F1-score of 99.85% and area under the curve (AUC) of 0.99 for multiclass classification. With its ability to differentiate between food intake and activity level, this device may potentially be used for tracking both energy intake and energy expenditure.