Supervised Machine Learning Classification Research Articles

Prediction of defensive success in elite soccer using machine learning - Tactical analysis of defensive play using tracking data and explainable AI

ABSTRACT The interest in sports performance analysis is rising and tracking data holds high potential for game analysis in team sports due to its accuracy and informative content. Together with machine learning approaches one can obtain deeper and more objective insights into the performance structure. In soccer, the analysis of the defense was neglected in comparison to the offense. Therefore, the aim of this study is to predict ball gains in defense using tracking data to identify tactical variables that drive defensive success. We evaluated tracking data of 153 games of German Bundesliga season 2020/21. With it, we derived player (defensive pressure, distance to the ball, & velocity) and team metrics (inter-line distances, numerical superiority, surface area, & spread) each containing a tactical idea. Afterwards, we trained supervised machine learning classifiers (logistic regression, XGBoost, & Random Forest Classifier) to predict successful (ball gain) vs. unsuccessful defensive plays (no ball gain). The expert-reduction-model (Random Forest Classifier with 16 features) showed the best and satisfying prediction performance (F1-Score (test) = 0.57). Analyzing the most important input features of this model, we are able to identify tactical principles of defensive play that appear to be related to gaining the ball: press the ball leading player, create numerical superiority in areas close to the ball (press short pass options), compact organization of defending team. Those principles are highly interesting for practitioners to gain valuable insights in the tactical behavior of soccer players that may be related to the success of defensive play.

Combination of personalized computational modeling and machine learning for optimization of left ventricular pacing site in cardiac resynchronization therapy.

Introduction: The 30-50% non-response rate to cardiac resynchronization therapy (CRT) calls for improved patient selection and optimized pacing lead placement. The study aimed to develop a novel technique using patient-specific cardiac models and machine learning (ML) to predict an optimal left ventricular (LV) pacing site (ML-PS) that maximizes the likelihood of LV ejection fraction (LVEF) improvement in a given CRT candidate. To validate the approach, we evaluated whether the distance DPS between the clinical LV pacing site (ref-PS) and ML-PS is associated with improved response rate and magnitude. Materials and methods: We reviewed retrospective data for 57 CRT recipients. A positive response was defined as a more than 10% LVEF improvement. Personalized models of ventricular activation and ECG were created from MRI and CT images. The characteristics of ventricular activation during intrinsic rhythm and biventricular (BiV) pacing with ref-PS were derived from the models and used in combination with clinical data to train supervised ML classifiers. The best logistic regression model classified CRT responders with a high accuracy of 0.77 (ROC AUC = 0.84). The LR classifier, model simulations and Bayesian optimization with Gaussian process regression were combined to identify an optimal ML-PS that maximizes the ML-score of CRT response over the LV surface in each patient. Results: The optimal ML-PS improved the ML-score by 17 ± 14% over the ref-PS. Twenty percent of the non-responders were reclassified as positive at ML-PS. Selection of positive patients with a max ML-score >0.5 demonstrated an improved clinical response rate. The distance DPS was shorter in the responders. The max ML-score and DPS were found to be strong predictors of CRT response (ROC AUC = 0.85). In the group with max ML-score > 0.5 and DPS< 30 mm, the response rate was 83% compared to 14% in the rest of the cohort. LVEF improvement in this group was higher than in the other patients (16 ± 8% vs. 7 ± 8%). Conclusion: A new technique combining clinical data, personalized heart modelling and supervised ML demonstrates the potential for use in clinical practice to assist in optimizing patient selection and predicting optimal LV pacing lead position in HF candidates for CRT.

Predictive analytics in smart healthcare for child mortality prediction using a machine learning approach.

In developing countries, child health and restraining under-five child mortality are one of the fundamental concerns. UNICEF adopted sustainable development goal 3 (SDG3) to reduce the under-five child mortality rate globally to 25 deaths per 1,000 live births. The under-five mortality rate is 69 deaths per 1,000 live child-births in Pakistan as reported by the Demographic and Health Survey (2018). Predictive analytics has the power to transform the healthcare industry, personalizing care for every individual. Pakistan Demographic Health Survey (2017-2018), the publicly available dataset, is used in this study and multiple imputation methods are adopted for the treatment of missing values. The information gain, a feature selection method, ranked the information-rich features and examine their impact on child mortality prediction. The synthetic minority over-sampling method (SMOTE) balanced the training dataset, and four supervised machine learning classifiers have been used, namely the decision tree classifier, random forest classifier, naive Bayes classifier, and extreme gradient boosting classifier. For comparative analysis, accuracy, precision, recall, and F1-score have been used. Eventually, a predictive analytics framework is built that predicts whether the child is alive or dead. The number under-five children in a household, preceding birth interval, family members, mother age, age of mother at first birth, antenatal care visits, breastfeeding, child size at birth, and place of delivery were found to be critical risk factors for child mortality. The random forest classifier performed efficiently and predicted under-five child mortality with accuracy (93.8%), precision (0.964), recall (0.971), and F1-score (0.967). The findings could greatly assist child health intervention programs in decision-making.

Evaluating expressway traffic crash severity by using logistic regression and explainable & supervised machine learning classifiers

The number of expressway road accidents in Sri Lanka has significantly increased (by 20%) due to the expansion of the transport network and high traffic volume. It is crucial to identify the causes of these crashes for effective road safety management. However, traditional statistical methods may be insufficient due to their inherent assumptions. This study utilized explainable machine learning to investigate the factors that affect the severity of traffic crashes on expressways. The study evaluated two groups of traffic crashes: fatal or severe crashes, and other crashes that included non-severe injuries or only property damage. Five factors that contribute to crashes were analyzed: road surface condition, road alignment, location, weather condition, and lighting effect. Four machine learning models (Random Forest (RF), Decision Tree (DT), extreme gradient boosting (XGB), K-Nearest Neighbor (KNN)) were developed and compared with Logistic Regression (LR) using 223 training and 56 testing data instances. The study revealed that the machine learning algorithms provided more accurate predictions than the LR model. To explain the machine learning models, Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) were used. These methods revealed that all five features decreased the possibility of occurrence of fatal accidents. SHAP and LIME explanations confirmed the known interactions between factors influencing crash severity in expressway operational conditions. These explanations increase the trust of end-users and domain experts on machine learning models. Furthermore, the study concluded that using explainable machine learning methods is more effective than traditional regression analysis in evaluating safety performance. Additionally, the results of the study can be utilized to improve road safety by providing accurate explanations for decision-making processes for black-box models.

Integrated failure analysis using machine learning predictive system for smart management of medical equipment maintenance

The study aims to develop the failure analysis predictive models, which prognosticate first failure event (FFE), failure-to-year ratio (FYR), and failure rectification group (FRG). The construction of predictive models involved nineteen categories of 13,350 units of medical equipment. We proposed thirteen novel features in assessing medical equipment failures. The failure analysis predictive models were categorised into several classes for training and testing the developed models. There was seven supervised machine learning classifiers and performance metrics applied in the experiment. The experiment demonstrates that Support Vector Machine is the best classifier for the FFE predictive model, which achieves an accuracy of 96.9% after hyperparameter optimisation. Furthermore, Decision Tree is the best classifier for FYR, with an accuracy of 83.9%. Meanwhile, the comparative analysis for FRG discovered that Artificial Neural Network achieved the highest accuracy among others with 76.7% accuracy after the hyperparameter optimisation process. Findings from this study indicate that this failure analysis predictive model functions as a main instrument for conducting predictive maintenance in the direction of smart maintenance practices. Through the developed predictive systems, timely maintenance of medical equipment can be performed. This will also assist healthcare service providers in initiating the remanufacturing and refurbishment programme, ensuring efficient medical care delivery. The suggested framework of machine learning-assisted failure analysis for medical equipment maintenance management may provide clinical engineers with guidance for managing the strategic maintenance management for medical equipment.

Navigating nature’s toll: Assessing the ecological impact of the refugee crisis in Cox’s Bazar, Bangladesh

The Rohingya crisis in Myanmar’s Rakhine state has resulted in a significant influx of refugees into Cox’s Bazar, Bangladesh. However, the ecological impact of this migration has received limited attention in research. This study aimed to address this gap by utilizing remote sensing data and machine learning techniques to model the ecological quality (EQ) of the region before and after the refugee influx. To quantify changes in land use and land cover (LULC), three supervised machine learning classification methods, namely artificial neural networks (ANN), support vector machines (SVM), and random forests (RF), were applied. The most accurate LULC maps obtained from these methods were then used to assess changes in ecosystem service valuation and function resulting from the land use changes. Furthermore, fuzzy logic models were employed to examine the EQ conditions before and after the Rohingya influx. The findings of the study indicate that the increased number of Rohingya refugees has led to a 9.58% decrease in forest area, accompanied by an 8.25% increase in settlement areas. The estimated total ecosystem services value (ESV) in the research area was $67.83 million in 2017 and $67.78 million in 2021, respectively. The ESV for forests experienced a significant decline of 21.97%, equivalent to a decrease of $5.33 million. Additionally, the reduction in forest lands has contributed to a 13.58% decline in raw materials and a 14.57% decline in biodiversity. Furthermore, utilizing a Markovian transition probability model, our analysis reveals that the EQ conditions in the area have deteriorated from “very good” or “good” to “bad” or “very bad” following the Rohingya influx. The findings of this study emphasize the importance of integrating ecological considerations into decision-making processes and developing proactive measures to mitigate the environmental impact of such large-scale migrations.

The PARTYPRESS Database: A new comparative database of parties’ press releases

We present the PARTYPRESS Database, which compiles more than 250,000 published press releases from 68 parties in 9 European countries. The database covers the press releases of the most relevant political parties in these countries from 2010 onward. It provides a supervised machine learning classification of press releases into 21 unique issue categories according to a general codebook. The PARTYPRESS Database can be used to study parties’ issue agendas comparatively and over time. We extend a recent analysis in Gessler and Hunger (2022) to illustrate the usefulness of the database in studying dynamic party competition, communication, and behavior.

O-187 The combination of machine learning and pharmacogenetics for the prediction of sub-optimal ovarian response

Abstract Study question Can an artificial intelligence (AI)-based model predict the risk of suboptimal ovarian response using patient genetic data? Summary answer AI-based predictive models can identify relevant predicting factors for suboptimal ovarian response, including several genetic variants. What is known already A substantial proportion of women classified as “normo-responders” may not reach the optimal range of oocytes retrieved after COS. Polyzos and Sunkara (2015) defined a new category of women (15-36%) with a worse prognosis and a “suboptimal” level of ovarian response in which the number of oocytes retrieved ranges between 4 and 9 oocytes. The explanation for this behavior has been suggested to be genetic and indeed variants in FSH receptor and LH subunit-B have been identified that predispose to lower ovarian sensitivity to stimulation. Therefore, these women might require higher doses of gonadotropins or longer stimulations. Study design, size, duration This observational study included a retrospective analysis of 1370 ovarian stimulations from egg donors performed between March 2018 and April 2022. The oocyte donor candidates were selected according to our clinic donation program requirements and ASRM and ESHRE guidelines for oocyte donation. They were stimulated following a progesterone-primed ovarian stimulation protocol. All donors started stimulation with 150-300 IU/day of FSH according to AFC and BMI. A GnRH agonist was used for final oocyte maturation. Participants/materials, setting, methods Oocyte donors were healthy women between 18 and 35 years old (n = 504). In order to establish the predictive machine learning models, patient characteristics and controlled ovarian stimulation data were recorded in a data frame. In addition, donors were genotyped for 31 variants corresponding to 16 genes associated with ovarian response. The association between the different variables and risk of suboptimal ovarian response was analysed using SPSS (v23.0) and R (v. 4.2.0) statistical software. Main results and the role of chance The oocyte donors had a mean age of 25.4±4.0 and a high ovarian reserve (AFC: 18.4±7). The mean number of oocytes retrieved after ovarian stimulation was 16.3±7.5, of which 13.1±6.4 were mature. Despite being young women with good ovarian reserve, 16.4% of the stimulations were suboptimal (4-9 oocytes retrieved). Classical statistical method (binary logistic regression) and 5 different supervised classification machine learning algorithms (multi-layer perceptron, support vector-machines, k-nearst neighbors, random forest and eXtreme Gradient Boosting (XGBoost)) were used to establish a prediction model for suboptimal ovarian response. The model with the highest AUC value was the XGBoost (0.876). The accuracy, sensitivity and specificity were 0.798, 0.809 and 0.787 respectively. The variables that had the greatest predictive power in the best machine learning algorithm (XGBoost) were: age, BMI, AFC and total gonadotrophin dose. In addition, four genetic variants were identified that modified the risk of a suboptimal ovarian response: -ESR2 (oestrogen receptor; c.*39G&amp;gt;A). -LHB (LH beta subunit; c.82T&amp;gt;C/p.Trp28Arg). -SOD2 (superoxide dismutase; c.47T&amp;gt;C/p.Val16Ala). A mitochondrial enzyme that catalyses the detoxification and protection against redox damage that can occur in COS. -and TP53 (tumour suppressor protein; c.215C&amp;gt;T/p.Pro72Leu) that exerts a protective effect on DNA damage in folliculogenesis. Limitations, reasons for caution Our investigation was performed with retrospectively collected data, and hence it will be of importance to collect data prospectively to confirm that the new identified variants are associated with suboptimal ovarian response also in the IVF population. Wider implications of the findings The combination of AI and pharmacogenetics has led to the identification of genetic variants that might predispose to suboptimal ovarian response. Women who carry these variants (erroneously considered as normo-responders) could be candidates for personalization of their ovarian stimulation treatment with higher doses of gonadotrophins and/or longer stimulations. Trial registration number Not applicable

Automated Classification of the Phases Relevant to Work-Related Musculoskeletal Injury Risks in Residential Roof Shingle Installation Operations Using Machine Learning

Awkward kneeling in sloped shingle installation operations exposes roofers to knee musculoskeletal disorder (MSD) risks. To address the varying levels of risk associated with different phases of shingle installation, this research investigated utilizing machine learning to automatically classify seven distinct phases in a typical shingle installation task. The classification process relied on analyzing knee kinematics data and roof slope information. Nine participants were recruited and performed simulated shingle installation tasks while kneeling on a sloped wooden platform. The knee kinematics data were collected using an optical motion capture system. Three supervised machine learning classification methods (i.e., k-nearest neighbors (KNNs), decision tree (DT), and random forest (RF)) were selected for evaluation. The KNN classifier provided the best performance for overall accuracy. The results substantiated the feasibility of applying machine learning in classifying shingle installation phases from workers’ knee joint rotation and roof slope angles, which may help facilitate method and tool development for automated knee MSD risk surveillance and assessment among roofers.

Synthetic Patient Data Generation and Evaluation in Disease Prediction Using Small and Imbalanced Datasets

The increasing prevalence of chronic non-communicable diseases makes it a priority to develop tools for enhancing their management. On this matter, Artificial Intelligence algorithms have proven to be successful in early diagnosis, prediction and analysis in the medical field. Nonetheless, two main issues arise when dealing with medical data: lack of high-fidelity datasets and maintenance of patient's privacy. To face these problems, different techniques of synthetic data generation have emerged as a possible solution. In this work, a framework based on synthetic data generation algorithms was developed. Eight medical datasets containing tabular data were used to test this framework. Three different statistical metrics were used to analyze the preservation of synthetic data integrity and six different synthetic data generation sizes were tested. Besides, the generated synthetic datasets were used to train four different supervised Machine Learning classifiers alone, and also combined with the real data. F1-score was used to evaluate classification performance. The main goal of this work is to assess the feasibility of the use of synthetic data generation in medical data in two ways: preservation of data integrity and maintenance of classification performance.

Using Natural Language Processing and Machine Learning to Identify Opioids in Electronic Health Record Data.

This study evaluates the utility of machine learning (ML) and natural language processing (NLP) in the processing and initial analysis of data within the electronic health record (EHR). We present and evaluate a method to classify medication names as either opioids or non-opioids using ML and NLP. A total of 4216 distinct medication entries were obtained from the EHR and were initially labeled by human reviewers as opioid or non-opioid medications. An approach incorporating bag-of-words NLP and supervised ML classification was implemented in MATLAB and used to automatically classify medications. The automated method was trained on 60% of the input data, evaluated on the remaining 40%, and compared to manual classification results. A total of 3991 medication strings were classified as non-opioid medications (94.7%), and 225 were classified as opioid medications by the human reviewers (5.3%). The algorithm achieved a 99.6% accuracy, 97.8% sensitivity, 94.6% positive predictive value, F1 value of 0.96, and a receiver operating characteristic (ROC) curve with 0.998 area under the curve (AUC). A secondary analysis indicated that approximately 15-20 opioids (and 80-100 non-opioids) were needed to achieve accuracy, sensitivity, and AUC values of above 90-95%. The automated approach achieved excellent performance in classifying opioids or non-opioids, even with a practical number of human reviewed training examples. This will allow a significant reduction in manual chart review and improve data structuring for retrospective analyses in pain studies. The approach may also be adapted to further analysis and predictive analytics of EHR and other "big data" studies.

The effect of seasonality in predicting the level of crime. A spatial perspective.

This paper presents an innovative methodology to study the application of seasonality (the existence of cyclical patterns) to help predict the level of crime. This methodology combines the simplicity of entropy-based metrics that describe temporal patterns of a phenomenon, on the one hand, and the predictive power of machine learning on the other. First, the classical Colwell's metrics Predictability and Contingency are used to measure different aspects of seasonality in a geographical unit. Second, if those metrics turn out to be significantly different from zero, supervised machine learning classification algorithms are built, validated and compared, to predict the level of crime based on the time unit. The methodology is applied to a case study in Barcelona (Spain), with month as the unit of time, and municipal district as the geographical unit, the city being divided into 10 of them, from a set of property crime data covering the period 2010-2018. The results show that (a) Colwell's metrics are significantly different from zero in all municipal districts, (b) the month of the year is a good predictor of the level of crime, and (c) Naive Bayes is the most competitive classifier, among those who have been tested. The districts can be ordered using the Naive Bayes, based on the strength of the month as a predictor for each of them. Surprisingly, this order coincides with that obtained using Contingency. This fact is very revealing, given the apparent disconnection between entropy-based metrics and machine learning classifiers.

Hybrid of Supervised Learning and Optimization Algorithm for Optimal Detection of IoT Distributed Denial of Service Attacks

The high-speed internet has led to the development of Internet of Things (IoT) with a fundamental Three-Layer IoT architecture. However, small amount of un-indicative data captured at the end level of IoT network makes the edge IoT devices susceptible to cyber-security attacks aimed at its transport layer. The Distributed Denial of Service (DDoS) poses significant cyber-security threat to the heterogenous IoT devices which are rendered vulnerable by ineffectiveness of conventional cybersecurity softwares. The literature reveals numerous studies that employed machine learning for the mitigation of IoT DDoS attacks but they lack in terms of an extensive investigation on optimization of machine learning classifiers. Therefore, this study first evaluates the prediction performance of machine learning classification algorithms trained on an authenticated/validated real-time IoT traffic dataset. The results reveal Logistic Regression (LR) as the most effective supervised machine learning classifier for detecting IoT DDoS attacks with a prediction accuracy of 97%. Following this, another investigation on the hybridization of LR with optimization algorithms yields Grasshopper Optimizer Algorithms (GOA) as the most effective optimizer in improving its prediction accuracy to 99%. Hence, the LR hybridized by GOA is developed as the optimal IoT DDoS Attack detection solution. Thus, the study serves to lay the foundation of a data-driven approach for the mitigation of the emerging variants of malicious IoT DDoS attacks such as zero-day attacks.

Event recognition in marine seismological data using Random Forest machine learning classifier

SUMMARYAutomatic detection of seismic events in ocean bottom seismometer (OBS) data is difficult due to elevated levels of noise compared to the recordings from land. Popular deep-learning approaches that work well with earthquakes recorded on land perform poorly in a marine setting. Their adaptation to OBS data requires catalogues containing hundreds of thousands of labelled event examples that currently do not exist, especially for signals different than earthquakes. Therefore, the usual routine involves standard amplitude-based detection methods and manual processing to obtain events of interest. We present here the first attempt to utilize a Random Forest supervised machine learning classifier on marine seismological data to automate catalogue screening and event recognition among different signals [i.e. earthquakes, short duration events (SDE) and marine noise sources]. The detection approach uses the short-term average/long-term average method, enhanced by a kurtosis-based picker for a more precise recognition of the onset of events. The subsequent machine learning method uses a previously published set of signal features (waveform-, frequency- and spectrum-based), applied successfully in recognition of different classes of events in land seismological data. Our workflow uses a small subset of manually selected signals for the initial training procedure and we then iteratively evaluate and refine the model using subsequent OBS stations within one single deployment in the eastern Fram Strait, between Greenland and Svalbard. We find that the used set of features is well suited for the discrimination of different classes of events during the training step. During the manual verification of the automatic detection results, we find that the produced catalogue of earthquakes contains a large number of noise examples, but almost all events of interest are properly captured. By providing increasingly larger sets of noise examples we see an improvement in the quality of the obtained catalogues. Our final model reaches an average accuracy of 87 per cent in recognition between the classes, comparable to classification results for data from land. We find that, from the used set of features, the most important in separating the different classes of events are related to the kurtosis of the envelope of the signal in different frequencies, the frequency with the highest energy and overall signal duration. We illustrate the implementation of the approach by using the temporal and spatial distribution of SDEs as a case study. We used recordings from six OBSs deployed between 2019 and 2020 off the west-Svalbard coast to investigate the potential link of SDEs to fluid dynamics and discuss the robustness of the approach by analysing SDE intensity, periodicity and distance to seepage sites in relation to other published studies on SDEs.

Improving Crop Mapping by Using Bidirectional Reflectance Distribution Function (BRDF) Signatures with Google Earth Engine

Recent studies have demonstrated the potential of using bidirectional reflectance distribution function (BRDF) signatures captured by multi-angle observation data to enhance land cover classification and retrieve vegetation architectures. Considering the diversity of crop architectures, we proposed that crop mapping precision may be enhanced by using BRDF signatures. We compared the accuracy of four supervised machine learning classifiers provided by the Google Earth Engine (GEE), namely random forest (RF), classification and regression trees (CART), support vector machine (SVM), and Naïve Bayes (NB), using the moderate resolution imaging spectroradiometer (MODIS) nadir BRDF-adjusted reflectance data (MCD43A4 V6) and BRDF and albedo model parameter data (MCD43A1 V6) as input. Our results indicated that using BRDF signatures leads to a moderate improvement in classification results in most cases, compared to using reflectance data from a single nadir observation direction. Specifically, the overall validation accuracy increased by up to 4.9%, and the validation kappa coefficients increased by up to 0.092. Furthermore, the classifiers were ranked in order of accuracy, from highest to lowest: RF, CART, SVM, and NB. Our study contributes to the development of crop mapping and the application of multi-angle observation satellites.

Integration of Per- and Polyfluoroalkyl Substance (PFAS) Fingerprints in Fish with Machine Learning for PFAS Source Tracking in Surface Water

Per- and polyfluoroalkyl substances (PFASs) are a class of environmental contaminants that originate from various sources. The unique chemical fingerprints associated with many commercial products and industrial applications make PFASs ideal candidates for machine learning (ML)-assisted environmental forensics. Here, we propose a novel use of PFAS fingerprints in fish tissue from surface water systems to classify exposure from multiple sources of PFASs using a proof-of-concept demonstration. Three supervised ML classification techniques (k-nearest neighbors (KNN), decision trees, support vector machines) implementing two predictive features are used to classify literature-reported PFAS fingerprints in fish (n = 1057). The importance of additional predictive features was explored using brute force optimization of a multifeature KNN algorithm. The multiclass classification considered exposure to aqueous film-forming foam-impacted water, paper industry wastewater, diffuse sources, or PFASs undergoing long-range transport. The optimized classifiers demonstrated 85%–94% classification accuracy for this first known multiclass classification of PFASs for environmental forensics. The optimized classifiers also demonstrated 79%–92% classification accuracy with a set of independent external validation data (n = 192). Our results demonstrate that PFAS fingerprints in fish tissue may be an effective means of PFAS source tracking in surface water systems. The source code is provided for guidance on best practices for ML-assisted environmental forensics.

Automatic Authorship Classification for German Lyrics Using Naïve Bayes

Text classification is a prevalent and essential machine-learning task. Machine learning classifiers have developed immensely since their inception. The naïve Bayes classifier is one of the most prominent supervised machine learning classifiers. In this experiment, we highlight the performance of Naïve Bayes for classifying of authors/artists on the German lyrics corpus (“Songkorpus”) and compare the classification results with other classifier algorithms. The corpus of investigation consists of six artists with 970 songs in total. Bayes model evaluation measures revealed a precision of 0.91, recall of 0.94, and F1-measure of 0.9. Furthermore, the classification performance with other classifier algorithms did not reveal any statistically significant difference in performance. The results of the study add to the high volume of reports on the classification accuracy of Naive Bayes for the task of lyrical classification.

Comparison between Machine Learning and Deep Learning Approaches for the Detection of Toxic Comments on Social Networks

The way we communicate has been revolutionised by the widespread use of social networks. Any kind of online message can reach anyone in the world almost instantly. The speed with which information spreads is undoubtedly the strength of social networks, but at the same time, any user of these platforms can see how toxic messages spread in parallel with likes, comments and ratings about any person or entity. In such cases, the victim feels even more helpless and defenceless as a result of the rapid spread. For this reason, we have implemented an automatic detector of toxic messages on social media. This allows us to stop toxicity in its tracks and protect victims. In particular, the aim of the survey is to demonstrate how traditional Machine Learning methods of Natural Language Processing (NLP) work on equal terms with Deep Learning methods represented by a Transformer architecture and characterised by a higher computational cost. In particular, the paper describes the results obtained by testing different supervised Machine Learning classifiers (Logistic Regression, Random Forest and Support Vector Machine) combined with two topic-modelling techniques of NLP, (Latent Semantic Analysis and Latent Dirichlet Allocation). A pre-trained Transformer named BERTweet was also tested. All models performed well in this task, so much so that values close to or above 90% were achieved in terms of the F1 score evaluation metric. The best result achieved by Transformer BERTweet, 91.40%, was therefore not impressive in this context, as the performance gains are too small compared to the computational overhead.

A Bearing Fault Diagnosis Method Based on Wavelet Denoising and Machine Learning

There are a lot of interference factors in the operating environment of machinery, which makes it ineffective to use traditional detection methods to judge the fault location and type of fault of the machinery, and even misjudgment of the fault location and type may occur. In order to solve these problems, this paper proposes a bearing fault diagnosis method based on wavelet denoising and machine learning. We use sensors to detect the operating conditions of rolling bearings under different working conditions to obtain datasets of different types of bearing failures. On the basis of using the wavelet denoising algorithm to reduce noise, we comprehensively evaluated five machine learning models, including K-means clustering, decision tree, random forest, and support vector machine to classify bearing faults and compare their results. By designing the fault classification evaluation prediction criteria, the following conclusions are drawn. The model proposed in this paper is significantly better than other traditional diagnostic models for bearing faults. In order to solve the problem of weak signal strength and background noise interference, this paper selects a better noise reduction algorithm under different quantitative evaluation indicators for wavelet denoising, which can better restore the true characteristics of the fault signal. Using unsupervised learning and supervised machine learning classification algorithms, the evaluation indicators before and after denoising are compared to make the classification results more accurate and reliable. This article will help researchers to intelligently diagnose the faults of rolling bearing equipment in rotating machinery.

A Multi-Feature Fusion Framework for Automatic Skin Cancer Diagnostics

Malignant melanoma is the most invasive skin cancer and is currently regarded as one of the deadliest disorders; however, it can be cured more successfully if detected and treated early. Recently, CAD (computer-aided diagnosis) systems have emerged as a powerful alternative tool for the automatic detection and categorization of skin lesions, such as malignant melanoma or benign nevus, in given dermoscopy images. In this paper, we propose an integrated CAD framework for rapid and accurate melanoma detection in dermoscopy images. Initially, an input dermoscopy image is pre-processed by using a median filter and bottom-hat filtering for noise reduction, artifact removal, and, thus, enhancing the image quality. After this, each skin lesion is described by an effective skin lesion descriptor with high discrimination and descriptiveness capabilities, which is constructed by calculating the HOG (Histogram of Oriented Gradient) and LBP (Local Binary Patterns) and their extensions. After feature selection, the lesion descriptors are fed into three supervised machine learning classification models, namely SVM (Support Vector Machine), kNN (k-Nearest Neighbors), and GAB (Gentle AdaBoost), to diagnostically classify melanocytic skin lesions into one of two diagnostic categories, melanoma or nevus. Experimental results achieved using 10-fold cross-validation on the publicly available MED-NODEE dermoscopy image dataset demonstrate that the proposed CAD framework performs either competitively or superiorly to several state-of-the-art methods with stronger training settings in relation to various diagnostic metrics, such as accuracy (94%), specificity (92%), and sensitivity (100%).