Accuracy Of Classifier Research Articles

Frequency-focused sound data generator for fault diagnosis in industrial robots

Abstract A frequency-focused sound data generator was developed for the in situ fault sound diagnosis of industrial robot reducers. The sound data generator, based on a conditional generative adversarial network, selects a target frequency range without relying on domain knowledge. A sound dataset of normal and faulty harmonic drive rotations of in situ industrial robots was collected using an attachable wireless sound sensor. The generated sound data were evaluated based on the fault diagnosis accuracy of a simple classifier trained using the generated data and tested using real data. The proposed method well-defined the frequency feature clusters and produced high-quality data, exhibiting up to 16.0% higher precision score on normal and 13.0% higher accuracy on weak-fault harmonic drive compared with the conventional methods, achieving fault diagnosis accuracy of 95.6% even in situations of fault data comprising only 5% of the normal data.

Double vision: 2D and 3D mosquito trajectories can be as valuable for behaviour analysis via machine learning

BackgroundMosquitoes are carriers of tropical diseases, thus demanding a comprehensive understanding of their behaviour to devise effective disease control strategies. In this article we show that machine learning can provide a performance assessment of 2D and 3D machine vision techniques and thereby guide entomologists towards appropriate experimental approaches for behaviour assessment. Behaviours are best characterised via tracking—giving a full time series of information. However, tracking systems vary in complexity. Single-camera imaging yields two-component position data which generally are a function of all three orthogonal components due to perspective; however, a telecentric imaging setup gives constant magnification with respect to depth and thereby measures two orthogonal position components. Multi-camera or holographic techniques quantify all three components.MethodsIn this study a 3D mosquito mating swarm dataset was used to generate equivalent 2D data via telecentric imaging and a single camera at various imaging distances. The performance of the tracking systems was assessed through an established machine learning classifier that differentiates male and non-male mosquito tracks. SHAPs analysis has been used to explore the trajectory feature values for each model.ResultsThe results reveal that both telecentric and single-camera models, when placed at large distances from the flying mosquitoes, can produce equivalent accuracy from a classifier as well as preserve characteristic features without resorting to more complex 3D tracking techniques.ConclusionsCaution should be exercised when employing a single camera at short distances as classifier balanced accuracy is reduced compared to that from 3D or telecentric imaging; the trajectory features also deviate compared to those from the other datasets. It is postulated that measurement of two orthogonal motion components is necessary to optimise the accuracy of machine learning classifiers based on trajectory data. The study increases the evidence base for using machine learning to determine behaviours from insect trajectory data.Graphical

An Approach for Efficient and Accurate Phishing Website Prediction Using Improved ML Classifier Performance for Feature Selection

The article discusses the use of machine learning (ML) to combat phishing websites, which are deceptive sites that mimic trusted entities to steal sensitive information. This is why the continued invention of methods of identifying and counteracting phishing threats is beneficial. Such attacks pose significant risks to the integrity of online security. To enhance the success rate and specificity of predicting phishing websites, this study proposes a new approach that utilizes machine learning algorithms. To enhance the methods mentioned above and achieve better results in classification and better prediction of customer behaviour, the main points exposed to further transformations are increasing classifier accuracy and selecting an optimal feature space. Traditional anti-phishing strategies like blacklisting and heuristic searches often have slow detection times and high false positive rates. The article introduces a novel feature selection method to extract highly correlated features from datasets, thereby enhancing classifier accuracy. Using six feature selection techniques on a phishing dataset, it evaluates eight classifiers, including SVM, Logistic Regression, Random Forest, and others. The study finds that the Random Forest classifier combined with the Chi-2 feature selection method significantly improves model accuracy, achieving up to 96.99%.

Comparative Analysis of SVM and CNN Techniques for Brain Tumor Detection

A brain tumor is the most common disease on earth and it is harmful to people. Tumors are the uncontrolled growth of cells and tissues in the human brain called a tumor. The image is acquired using CT scans and Magnetic Resonance Images. The identification of tumors at an early stage is critical and challenging for researchers. A patient comes to the hospital when he starts suffering from pain, headache, omission etc and at that time, if he has a tumor, To recognize the tumor early stage it is very different to identify whether it is benign (non-cancerous) or malignant (cancerous), many techniques or methods are available for detection of tumor here we apply SVM algorithm and CNN on brain Magnetic Resonance Images for classification of a benign or malignant tumor. Here, we propose a system based on the new concept of simple tumor detection that uses feature extraction techniques, segmentation algorithm and classification. To identify similar patients who have or do not have a brain tumor, as well as to ascertain the type of tumor they have and their tumor sizes. By comparing both SVM & CNN which technique is more beneficial and which one is better in both? The performance of SVM classifiers is measured in terms of training effectiveness and classification accuracy. With 95% accuracy, it manages the process of brain tumor categorization in MRI scans. The efficacy of training and classification accuracy of the CNN classifier is compared (96.33%). Both methods get high accuracy but as compared to SVM, CNN provides more accuracy and consumes less time for execution.

A New Alternating Suboptimal Dynamic Programming Algorithm with Applications for Feature Selection

Feature selection is predominantly used in machine learning tasks, such as classification, regression, and clustering. It selects a subset of features (relevant attributes of data points) from a larger set that contributes as optimally as possible to the informativeness of the model. There are exponentially many subsets of a given set, and thus, the exhaustive search approach is only practical for problems with at most a few dozen features. In the past, there have been attempts to reduce the search space using dynamic programming. However, models that consider similarity in pairs of features alongside the quality of individual features do not provide the required optimal substructure. As a result, algorithms, which we will call suboptimal dynamic programming algorithms, find a solution that may deviate significantly from the optimal one. In this paper, we propose an iterative dynamic programming algorithm, which invertsthe order of feature processing in each iteration. Such an alternating approach allows for improving the optimization function by using the score from the previous iteration to estimate the contribution of unprocessed features. The iterative process is proven to converge and terminates when the solution does not change in three successive iterations or when the number of iterations reaches the threshold. Results in more than 95% of tests align with those of the exhaustive search approach, being competitive and often superior to the reference greedy approach. Validation was carried out by comparing the scores of output feature subsets and examining the accuracy of different classifiers learned on these features across nine real-world applications, considering different scenarios with various numbers of features and samples. In the context of feature selection, the proposed algorithm can be characterized as a robust filter method that can improve machine learning models regardless of dataset size. However, we expect that the idea of alternating suboptimal optimization will soon be generalized to tasks beyond feature selection.

Machine learning and deep learning for classifying the justification of brain CT referrals.

To train the machine and deeplearning models to automate the justification analysis of radiology referrals in accordance with iGuide categorisation, and to determine if prediction models can generalise across multiple clinical sites and outperform human experts. Adult brain computed tomography (CT) referrals from scans performed in three CT centres in Ireland in 2020 and 2021 were retrospectively collected. Two radiographers analysed the justification of 3000 randomly selected referrals using iGuide, with two consultant radiologists analysing the referrals with disagreement. Insufficient or duplicate referrals were discarded. The inter-rater agreement among radiographers and consultants was computed. A random split (4:1) was performed to apply machine learning (ML) and deeplearning (DL) techniques to unstructured clinical indications to automate retrospective justification auditing with multi-class classification. The accuracy and macro-averaged F1 score of the best-performing classifier of each type on the training set were computed on the test set. 42 referrals were ignored. 1909 (64.5%) referrals were justified, 811 (27.4%) were potentially justified, and 238 (8.1%) were unjustified. The agreement between radiographers (κ = 0.268) was lower than radiologists (κ = 0.460). The best-performing ML model was the bag-of-words-based gradient-boosting classifier achieving a 94.4% accuracy and a macro F1 of 0.94. DL models were inferior, with bi-directional long short-term memory achieving 92.3% accuracy, a macro F1 of 0.92, and outperforming multilayer perceptrons. Interpreting unstructured clinical indications is challenging necessitating clinical decision support. ML and DL can generalise across multiple clinical sites, outperform human experts, and be used as an artificial intelligence-based iGuide interpreter when retrospectively vetting radiology referrals. Healthcare vendors and clinical sites should consider developing and utilising artificial intelligence-enabled systems for justifying medical exposures. This would enable better implementation of imaging referral guidelines in clinical practices and reduce population dose burden, CT waiting lists, and wasteful use of resources. Significant variations exist among human experts in interpreting unstructured clinical indications/patient presentations. Machine and deep learning can automate the justification analysis of radiology referrals according to iGuide categorisation. Machine and deep learning can improve retrospective and prospective justification auditingfor better implementation of imaging referral guidelines.

Feature selection considering feature relevance, redundancy and interactivity for neighbourhood decision systems

Feature selection is an effective method to simplify data analysis and obtain key features, which improves the accuracy and generalization ability of classifiers. Neighbourhood rough set is a typical granular computing model that enables data analysis at different granularity by setting different neighbourhood radii. In recent years, feature selection based on neighbourhood rough sets has received widespread attention from researchers. However, these models still have the following shortcomings. On the one hand, they rarely simultaneously consider the relevance, the redundancy and interactivity among features, making it difficult to accurately depict classification information in the data. On the other hand, there is still a lack of effective uncertainty measures to evaluate the classification ability of feature subsets. For these reasons, this article proposes a feature selection model that considers the relevance, redundancy, and interactivity of features. Firstly, the relevance, the redundancy and interactivity between features, are defined based on the neighbourhood, respectively. Secondly, a new feature evaluation function (MMM function) is proposed based on the principles of maximum relevance, minimum redundancy, and maximum interactivity. Finally, from the perspective of maintaining classification ability, a heuristic feature selection algorithm (FS_CRRI) was designed. To evaluate the performance of the proposed algorithm, it was compared with some representative feature selection algorithms on twelve common datasets. The experimental results on three different classifiers indicate that the FS_CRRI algorithm can effectively reduce data and achieve higher classification performance.

Simulation of artificial intelligence robots in dance action recognition and interaction process based on machine vision

Artificial intelligence robots based on machine vision have shown great potential in dance action recognition and interaction. In order to perform action recognition, a large amount of dance action video data was collected as the basis for training and testing the model. In motion recognition, various computer vision techniques are used to extract motion features from images or videos to capture key information of dance movements. Machine learning technology was adopted to construct an action classification model. By using the data collected in the early stage and the extracted features, an efficient and accurate classifier has been studied and trained. It can classify the input dance actions into different categories, segment continuous video frames into different action segments, and ensure accurate differentiation of different dance actions during the recognition process, avoiding misjudgment or confusion. By accurately capturing and analyzing the posture information of the human body, we can better understand and simulate dance movements, and enhance the performance and realism of artificial intelligence robots in dance interactions. The experimental results show that the system can accurately recognize various dance movements and achieve good interaction with users.

A Large Scale Study and Classification of VirusTotal Reports on Phishing and Malware URLs

VirusTotal (VT) is a widely used scanning service for researchers and practitioners to label malicious entities and predict new security threats. Unfortunately, it is little known to the end-users how VT URL scanners decide on the maliciousness of entities and the attack types they are involved in (e.g., phishing or malware-hosting websites). In this paper, we conduct a systematic comparative study on VT URL scanners' behavior for different attack types of malicious URLs, in terms of 1) detection specialties, 2) stability, 3) correlations between scanners, and 4) lead/lag behaviors. Our findings highlight that the VT scanners commonly disagree with each other on their detection and attack type classification, leading to challenges in ascertaining the maliciousness of a URL and taking prompt mitigation actions according to different attack types. This motivates us to present a new highly accurate classifier that helps correctly identify the attack types of malicious URLs at the early stage. This in turn assists practitioners in performing better threat aggregation and choosing proper mitigation actions for different attack types.

Diverse and tailored image generation for zero-shot multi-label classification

Recently, zero-shot multi-label classification has garnered considerable attention owing to its capacity to predict unseen labels without human annotations. Nevertheless, prevailing approaches often use seen classes as imperfect proxies for unseen classes, resulting in suboptimal performance. Drawing inspiration from the success of text-to-image generation models in producing realistic images, we propose an innovative solution: generating synthetic data to construct a training set explicitly tailored for proxyless training of unseen labels. Our approach introduces a novel image generation framework that produces multi-label synthetic images of unseen classes for classifier training. To enhance the diversity of the generated images, we leverage a pretrained large language model to generate diverse prompts. Employing a pretrained multimodal contrastive language-image pretraining (CLIP) model as a discriminator, we assessed whether the generated images accurately represented the target classes. This enables automatic filtering of inaccurately generated images and preserves classifier accuracy. To refine the text prompts for more precise and effective multi-label object generation, we introduced a CLIP score-based discriminative loss to fine-tune the text encoder in the diffusion model. In addition, to enhance the visual features of the target task while maintaining the generalization of the original features and mitigating catastrophic forgetting resulting from fine-tuning the entire visual encoder, we propose a feature fusion module inspired by transformer attention mechanisms. This module aids in capturing the global dependencies between multiple objects more effectively. Extensive experimental results validated the effectiveness of our approach, demonstrating significant improvements over state-of-the-art methods. The code is available at https://github.com/TAKELAMAG/Diff-ZS-MLC.

Machine learning assisted prediction of land surface temperature (LST) based on major air pollutants over the Annamayya District of India

Remote sensing (RS), Geographic information systems (GIS), and Machine learning can be integrated to predict land surface temperatures (LST) based on the data related to carbon monoxide (CO), Formaldehyde (HCHO), Nitrogen dioxide (NO2), Sulphur dioxide (SO2), absorbing aerosol index (AAI), and Aerosol optical depth (AOD). In this study, LST was predicted using machine learning classifiers, i.e., Extra trees classifier (ET), Logistic regressors (LR), and Random Forests (RF). The accuracy of the LR classifier (0.89 or 89%) is higher than ET (82%) and RF (82%) classifiers. Evaluation metrics for each classifier are presented in the form of accuracy, Area under the curve (AUC), Recall, Precision, F1 score, Kappa, and MCC (Matthew’s correlation coefficient). Based on the relative performance of the ML classifiers, it was concluded that the LR classifier performed better. Geographic information systems and RS tools were used to extract the data across spatial and temporal scales (2019 to 2022). In order to evaluate the model graphically, ROC (Receiver operating characteristic) curve, Confusion matrix, Validation curve, Classification report, Feature importance plot, and t- SNE (t-distributed stochastic neighbour embedding) plot were used. On validation of each ML classifier, it was observed that the RF classifier returned model complexity due to limited data availability and other factors yet to be studied post data availability. Sentinel-5-P and MODIS data are used in this study.

Feature Selection Using Hybrid Metaheuristic Algorithm for Email Spam Detection

Abstract In the present study, Krill Herd (KH) is proposed as a Feature Selection tool to detect spam email problems. This works by assessing the accuracy and performance of classifiers and minimizing the number of features. Krill Herd is a relatively new technique based on the herding behavior of small crustaceans called krill. This technique has been combined with a local search algorithm called Tabu Search (TS) and has been successfully employed to identify spam emails. This method has also generated much better results than other hybrid algorithm optimization systems such as the hybrid Water Cycle Algorithm with Simulated Annealing (WCASA). To assess the effectiveness of KH algorithms, SVM classifiers, and seven benchmark email datasets were used. The findings indicate that KHTS is much more accurate in detecting spam mail (97.8%) than WCASA.

Machine learning trial to detect sex differences in simple sticker arts of 1606 preschool children.

Previous studies suggested that drawings made by preschool boys and girls show distinguishable differences. However, children's drawings on their own are too complexly determined and inherently ambiguous to be a reliable indicator. In the present study, we attempted to develop a machine learning algorithm for classification of sex of the subjects using children's artworks. We studied three types of simple sticker artworks from 1606 Japanese preschool children aged 51-83 months (803 boys and 803 girls). Those artworks were processed into digitalized data. Simulated data based on the original data were also generated. Logistic regression approach was applied to each dataset to make a classifier, and run on each dataset in a stratified ten-fold cross-validation with hyperparameter tuning. A probability score was calculated in each sample and utilized for sex classification. Prediction performance was evaluated using accuracy, recall, and precision scores, as well as learning curves. Two models created from the original and simulated data showed comparably low metrics. The distributions of probability scores in the samples from boys and girls mostly overlapped and were indistinguishable. Learning curves of the models showed an extremely under-fitted pattern. Our machine learning algorithm was unable to distinguish simple sticker arts created by boys and girls. More complex tasks will enable to develop an accurate classifier.

A software defect prediction method using binary gray wolf optimizer and machine learning algorithms

ContextSoftware defect prediction means finding defect-prone modules before the testing process which will reduce testing cost and time. Machine learning methods can provide valuable models for developers to classify software faulty modules. ProblemThe inherent problem of the classification is the large volume of the training dataset's features, which reduces the accuracy and precision of the classification results. The selection of the effective features of the training dataset for classification is an NP-hard problem that can be solved using heuristic algorithms. MethodIn this study, a binary version of the Gray Wolf optimizer (bGWO) was developed to select the most effective features of the training dataset. By selecting the most influential features in the classification, the precision and accuracy of the software module classifiers can be increased. ContributionDeveloping a binary version of the gray wolf optimization algorithm to optimally select the effective features and creating an effective defect predictor are the main contributions of this study. To evaluate the effectiveness of the proposed method, five real-world and standard datasets have been used for the training and testing stages of the classifier. ResultsThe results indicate that among the 21 features of the train datasets, the basic complexity, sum of operators and operands, lines of codes, number of lines containing code and comments, and sum of operands have the greatest effect in predicting software defects. In this research, by combining the bGWO method and machine learning algorithms, accuracy, precision, recall, and F1 criteria have been considerably increased.

FetalBrainAwareNet: Bridging GANs with anatomical insight for fetal ultrasound brain plane synthesis

Over the past decade, deep-learning (DL) algorithms have become a promising tool to aid clinicians in identifying fetal head standard planes (FHSPs) during ultrasound (US) examination. However, the adoption of these algorithms in clinical settings is still hindered by the lack of large annotated datasets. To overcome this barrier, we introduce FetalBrainAwareNet, an innovative framework designed to synthesize anatomically accurate images of FHSPs. FetalBrainAwareNet introduces a cutting-edge approach that utilizes class activation maps as a prior in its conditional adversarial training process. This approach fosters the presence of the specific anatomical landmarks in the synthesized images. Additionally, we investigate specialized regularization terms within the adversarial training loss function to control the morphology of the fetal skull and foster the differentiation between the standard planes, ensuring that the synthetic images faithfully represent real US scans in both structure and overall appearance. The versatility of our FetalBrainAwareNet framework is highlighted by its ability to generate high-quality images of three predominant FHSPs using a singular, integrated framework. Quantitative (Fréchet inception distance of 88.52) and qualitative (t-SNE) results suggest that our framework generates US images with greater variability compared to state-of-the-art methods. By using the synthetic images generated with our framework, we increase the accuracy of FHSP classifiers by 3.2% compared to training the same classifiers solely with real acquisitions. These achievements suggest that using our synthetic images to increase the training set could provide benefits to enhance the performance of DL algorithms for FHSPs classification that could be integrated in real clinical scenarios.

Using machine learning algorithms to enhance IoT system security.

The term "Internet of Things" (IoT) refers to a system of networked computing devices that may work and communicate with one another without direct human intervention. It is one of the most exciting areas of computing nowadays, with its applications in multiple sectors like cities, homes, wearable equipment, critical infrastructure, hospitals, and transportation. The security issues surrounding IoT devices increase as they expand. To address these issues, this study presents a novel model for enhancing the security of IoT systems using machine learning (ML) classifiers. The proposed approach analyzes recent technologies, security, intelligent solutions, and vulnerabilities in ML IoT-based intelligent systems as an essential technology to improve IoT security. The study illustrates the benefits and limitations of applying ML in an IoT environment and provides a security model based on ML that manages autonomously the rising number of security issues related to the IoT domain. The paper proposes an ML-based security model that autonomously handles the growing number of security issues associated with the IoT domain. This research made a significant contribution by developing a cyberattack detection solution for IoT devices using ML. The study used seven ML algorithms to identify the most accurate classifiers for their AI-based reaction agent's implementation phase, which can identify attack activities and patterns in networks connected to the IoT. The study used seven ML algorithms to identify the most accurate classifiers for their AI-based reaction agent's implementation phase, which can identify attack activities and patterns in networks connected to the IoT. Compared to previous research, the proposed approach achieved a 99.9% accuracy, a 99.8% detection average, a 99.9 F1 score, and a perfect AUC score of 1. The study highlights that the proposed approach outperforms earlier machine learning-based models in terms of both execution speed and accuracy. The study illustrates that the suggested approach outperforms previous machine learning-based models in both execution time and accuracy.

GR.5 Establishing the utility of multi-platform liquid biopsy by integrating the CSF methylome and proteome in CNS malignancy

Background: Liquid biopsy represents a major development in cancer research, with significant translational potential. Similarly, the integration of multiple molecular platforms has yielded novel insights into disease biology and heterogeneity. We hypothesise that applying contemporary multi-omic approaches to liquid biopsies will improve the power of current models. Methods: We have compiled a cohort of 51 patients with glioblastoma, brain metastasis, and primary CNS lymphoma who underwent CSF sampling as part of clinical care. Cell free methylated DNA and shotgun proteomic profiling was obtained from the CSF of each patient and used to build tumour-specific classifiers. Integrated classifiers were compared with single platform classifiers using multiple approaches. Results: In this study, we show that the DNA methylation and protein profiles of cerebrospinal fluid can be combined to fully discriminate lymphomas from their major diagnostic counterparts with perfect AUCs of 1 (95% confidence interval 1-1) and 100% specificity. Each integrated lymphoma classifier significantly outperforms single-platform classifiers, suggesting synergistic biology is obtained using multiple molecular platforms. Conclusions: We present the most specific and accurate CNS lymphoma classifier to date by integrating the methylome and proteome of CSF. This has important implications for the future of cancer diagnostics and generates immediate utility for patients with CNS lymphoma.

Diagnostic accuracy of Afirma gene expression classifier, Afirma gene sequencing classifier, ThyroSeq v2 and ThyroSeq v3 for indeterminate (Bethesda III and IV) thyroid nodules: a meta-analysis.

The management of thyroid nodules with indeterminate cytology (ITN) is still a challenge. To evaluate the performance of commercial molecular tests for ITN, we performed this comprehensive meta-analysis. We performed an electronic search using PubMed/Medline, Embase, and the Cochrane Library. Studies assessing the diagnostic accuracy of Afirma gene expression classifier (GEC), Afirma gene sequencing classifier (GSC), ThyroSeq v2 (TSv2), or ThyroSeq v3 (TSv3) in patients with ITN (only Bethesda category III or IV) were selected; Statistical analyses were performed by using Stata. Seventy-one samples (GEC, n = 38; GSC, n = 16; TSv2, n = 9; TSv3, n = 8) in 53 studies, involving 6490 fine needle aspirations (FNAs) with ITN cytology with molecular diagnostics (GEC, GSC, TSv2, or TSv3), were included in the study. The meta-analysis showed the following pooled estimates: sensitivity 0.95 (95% CI: 0.94-0.97), specificity 0.35 (0.28-0.43), positive likelihood ratio (LR+) 1.5 (1.3-1.6), and negative likelihood ratio (LR-) 0.13 (0.09-0.19), with the best performance for TSv3 (area under the ROC curve 0.95 (0.93-0.96), followed by TSv2 (0.90 (0.87-0.92)), GSC (0.86 (0.82-0.88)), and GEC (0.82 (0.78-0.85)); the best rule-out property was observed for GSC (LR-, 0.07 (0.02-0.19)), followed by TSv3 (0.11 (0.05-0.24)) and GEC (0.16 (0.10-0.28), and the best rule-in was observed for TSv2 (LR+, 2,9 (1.4-4.6)), followed by GSC (1.9 (1.6-2.4)). A meta-regression analysis revealed that study design, Bethesda category, and type of molecular test were independent factors. We showed that in patients with ITN, TSv3 has the best molecular diagnostic performance, followed by TSv2, GSC, and GEC. As regards rule-out malignancy, GSC, and rule-in, TSV2 is superior to other tests.

Continuous and Unconstrained Tremor Monitoring in Parkinson's Disease Using Supervised Machine Learning and Wearable Sensors.

Accurately assessing the severity and frequency of fluctuating motor symptoms is important at all stages of Parkinson's disease management. Contrarily to time-consuming clinical testing or patient self-reporting with uncertain reliability, recordings with wearable sensors show promise as a tool for continuously and objectively assessing PD symptoms. While wearables-based clinical assessments during standardised and scripted tasks have been successfully implemented, assessments during unconstrained activity remain a challenge. We developed and implemented a supervised machine learning algorithm, trained and tested on tremor scores. We evaluated the algorithm on a 67-hour database comprising sensor data and clinical tremor scores for 24 Parkinson patients at four extremities for periods of about 3 hours. A random 25% subset of the labelled samples was used as test data, the remainder as training data. Based on features extracted from the sensor data, a Support Vector Machine was trained to predict tremor severity. Due to the inherent imbalance in tremor scores, we applied dataset rebalancing techniques. Our classifier demonstrated robust performance in detecting tremor events with a sensitivity of 0.90 on the test-portion of the resampled dataset. The overall classification accuracy was high at 0.88. We implemented an accurate classifier for tremor monitoring in free-living environments that can be trained even with modestly sized and imbalanced datasets. This advancement offers significant clinical value in continuously monitoring Parkinson's disease symptoms beyond the hospital setting, paving the way for personalized management of PD, timely therapeutic adjustments, and improved patient quality of life.

De Novo Antimicrobial Peptide Design with Feedback Generative Adversarial Networks.

Antimicrobial peptides (AMPs) are promising candidates for new antibiotics due to their broad-spectrum activity against pathogens and reduced susceptibility to resistance development. Deep-learning techniques, such as deep generative models, offer a promising avenue to expedite the discovery and optimization of AMPs. A remarkable example is the Feedback Generative Adversarial Network (FBGAN), a deep generative model that incorporates a classifier during its training phase. Our study aims to explore the impact of enhanced classifiers on the generative capabilities of FBGAN. To this end, we introduce two alternative classifiers for the FBGAN framework, both surpassing the accuracy of the original classifier. The first classifier utilizes the k-mers technique, while the second applies transfer learning from the large protein language model Evolutionary Scale Modeling 2 (ESM2). Integrating these classifiers into FBGAN not only yields notable performance enhancements compared to the original FBGAN but also enables the proposed generative models to achieve comparable or even superior performance to established methods such as AMPGAN and HydrAMP. This achievement underscores the effectiveness of leveraging advanced classifiers within the FBGAN framework, enhancing its computational robustness for AMP de novo design and making it comparable to existing literature.