Machine Learning Solutions Research Articles

CARMEL: Capturing spatio-temporal correlations via time-series sub-window imaging for home appliance classification

Energy management systems (EMS), as enablers of more efficient energy consumption, monitor and manage appliances to help residents be more energy efficient and thus more frugal. Recent appliance detection and identification techniques for such systems rely on machine learning. However, machine learning solutions for appliance classification on existing low-frequency household metering have not yet been thoroughly investigated. In this paper, we propose CARMEL, a new approach for identifying home appliances from load monitoring in building EMS based on a new data representation technique and a new model that leverages spatio-temporal correlations in the new representation. The proposed data representation technique performs dimensionality expansion of time series that scales linearly, rather than quadratically and, together with the proposed model, outperform the state of the art image transformation models by 5 percentage points. Evaluation on 5 different low-frequency household metering datasets, considering 29 appliances in total, shows that the proposed representation and the corresponding resource-aware deep learning architecture (1) achieve an average weighted F1 score of 0.92 and (2) require only 230 labeled samples and 3x fewer epochs to transfer to new households.

Uncertainty quantification in machine learning for engineering design and health prognostics: A tutorial

On top of machine learning (ML) models, uncertainty quantification (UQ) functions as an essential layer of safety assurance that could lead to more principled decision making by enabling sound risk assessment and management. The safety and reliability improvement of ML models empowered by UQ has the potential to significantly facilitate the broad adoption of ML solutions in high-stakes decision settings, such as healthcare, manufacturing, and aviation, to name a few. In this tutorial, we aim to provide a holistic lens on emerging UQ methods for ML models with a particular focus on neural networks and the applications of these UQ methods in tackling engineering design as well as prognostics and health management problems. Towards this goal, we start with a comprehensive classification of uncertainty types, sources, and causes pertaining to UQ of ML models. Next, we provide a tutorial-style description of several state-of-the-art UQ methods: Gaussian process regression, Bayesian neural network, neural network ensemble, and deterministic UQ methods focusing on spectral-normalized neural Gaussian process. Established upon the mathematical formulations, we subsequently examine the soundness of these UQ methods quantitatively and qualitatively (by a toy regression example) to examine their strengths and shortcomings from different dimensions. Then, we review quantitative metrics commonly used to assess the quality of predictive uncertainty in classification and regression problems. Afterward, we discuss the increasingly important role of UQ of ML models in solving challenging problems in engineering design and health prognostics. Two case studies with source codes available on GitHub are used to demonstrate these UQ methods and compare their performance in the life prediction of lithium-ion batteries at the early stage (case study 1) and the remaining useful life prediction of turbofan engines (case study 2).

Machine Learning Techniques for Sustainable Smart Cities Traffic Management

The issue of traffic congestion is one of the problems that disrupt the smoothness and comfort of the community in carrying out daily affairs. Rapid development, with an expanding population and vehicles in cities, has caused traffic congestion to become one of the most pressing concerns. As a result, this study proposed vehicle tracking method for traffic system management to improve and keep up with the rising demand for better traffic on the road. In the proposed system, cameras were placed at traffic intersections and along the road or highway to gather real-time traffic data. Image processing and machine learning algorithm were applied to calculate the number of vehicles based on the traffic flow and vehicle speed on the road to reflect the current traffic condition so that it helps general public to plan their trip and less pollution is produced in long run. This study focused on the cities of Cyberjaya, Putrajaya and selected Klang Valley areas to study the traffic condition during the peak and off-peak hour. Experimental results showed that the proposed machine learning solution provided high accuracy and effective detection on real-time traffic congestion. Current method used variables such as vehicle speed and number of vehicles to calculate the traffic flow on the road. As future direction, this method can be expanded to incorporate vehicle density as one of the variables for more precise traffic congestion reporting.

Automated Prediction of Crack Propagation Using H2O AutoML.

Crack propagation is a critical phenomenon in materials science and engineering, significantly impacting structural integrity, reliability, and safety across various applications. The accurate prediction of crack propagation behavior is paramount for ensuring the performance and durability of engineering components, as extensively explored in prior research. Nevertheless, there is a pressing demand for automated models capable of efficiently and precisely forecasting crack propagation. In this study, we address this need by developing a machine learning-based automated model using the powerful H2O library. This model aims to accurately predict crack propagation behavior in various materials by analyzing intricate crack patterns and delivering reliable predictions. To achieve this, we employed a comprehensive dataset derived from measured instances of crack propagation in Acrylonitrile Butadiene Styrene (ABS) specimens. Rigorous evaluation metrics, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R-squared (R2) values, were applied to assess the model's predictive accuracy. Cross-validation techniques were utilized to ensure its robustness and generalizability across diverse datasets. Our results underscore the automated model's remarkable accuracy and reliability in predicting crack propagation. This study not only highlights the immense potential of the H2O library as a valuable tool for structural health monitoring but also advocates for the broader adoption of Automated Machine Learning (AutoML) solutions in engineering applications. In addition to presenting these findings, we define H2O as a powerful machine learning library and AutoML as Automated Machine Learning to ensure clarity and understanding for readers unfamiliar with these terms. This research not only demonstrates the significance of AutoML in future-proofing our approach to structural integrity and safety but also emphasizes the need for comprehensive reporting and understanding in scientific discourse.

A Critical Evaluation of Business Improvement through Machine Learning: Challenges, Opportunities, and Best Practices

This paper presents a critical evaluation of the impact of machine learning (ML) on business improvement, focusing on the challenges, opportunities, and best practices associated with its implementation. The study examines the hurdles faced by businesses while integrating ML, such as data quality, talent acquisition, algorithm bias, interpretability, and privacy concerns. On the other hand, it highlights the advantages of ML, including data-driven decision-making, enhanced customer experience, process optimization, cost reduction, and the potential for new revenue streams. Furthermore, the paper offers best practices to guide businesses in successfully adopting ML solutions, covering data management, talent development, model evaluation, ethics, and regulatory compliance. Through real-world case studies, the study illustrates successful ML applications in different industries. It also addresses the ethical and social implications of ML adoption and discusses emerging trends for future directions. Ultimately, this evaluation provides valuable insights to enable informed decisions and sustainable growth for businesses leveraging machine learning.

Revolutionizing ransomware detection and criticality assessment: Multiclass hybrid machine learning and semantic similarity-based end2end solution

Revolutionizing ransomware detection and criticality assessment: Multiclass hybrid machine learning and semantic similarity-based end2end solution

A Comprehensive Review of the GNSS with IoT Applications and Their Use Cases with Special Emphasis on Machine Learning and Deep Learning Models

This paper presents a comprehensive review of the Global Navigation Satellite System (GNSS) with Internet of Things (IoT) applications and their use cases with special emphasis on Machine learning (ML) and Deep Learning (DL) models. Various factors like the availability of a huge amount of GNSS data due to the increasing number of interconnected devices having low-cost data storage and low-power processing technologies - which is majorly due to the evolution of IoT - have accelerated the use of machine learning and deep learning based algorithms in the GNSS community. IoT and GNSS technology can track almost any item possible. Smart cities are being developed with the use of GNSS and IoT. This survey paper primarily reviews several machine learning and deep learning algorithms and solutions applied to various GNSS use cases that are especially helpful in providing accurate and seamless navigation solutions in urban areas. Multipath, signal outages with less satellite visibility, and lost communication links are major challenges that hinder the navigation process in crowded areas like cities and dense forests. The advantages and disadvantages of using machine learning techniques are also highlighted along with their potential applications with GNSS and IoT.

Implementing integrated digital twin modelling and representation into the Snap4City platform for smart city solutions

Recently, Digital Twins solutions have attracted a growing interest as a fundamental paradigm for managing data driven processes on smart cities. They are complex modelling that should include 3D interactive representations of buildings and infrastructures, integrated with a wide range of data for Smart City cyber-physical ecosystem monitoring and controlling. This paper presents a framework for modelling, generating and distributing Digital Twin representations with 3D models from a various set of data, as well as its integration into the open-source Smart City framework, where many kinds of real time and historical data are available. The proposed solution offers a method for creating integrated data rendering of 3D city entities coupled with Smart City data (e.g., IoT Devices with time-series and historical data, heatmaps, geometries and shapes related to traffic flows, bus routes/stops, cycling paths). The solution for generating 3D representation is based on a number of computer vision and machine learning solutions, thus shortening the activities of passing from raw data (i.e., Lidar, shapes, patterns, etc.) to 3D representations. Implementation has been enforced into the quite widespread open-source Snap4City Smart City platform and has been validated by using hundreds of buildings in Florence city central area, Italy, plus hundreds of thousands of data as points of interest, IoT Devices, traffic flows, dynamic heatmaps, etc.

RETRACTED ARTICLE: Enhancing throughput using channel access priorities in frequency hopping network using federated learning

The data are sent by the nodes taking part in frequency hopping communications (FHC) utilising carrier frequencies and time slots that are pseudo-randomly assigned. Because of this, a high degree of protection against eavesdropping and anti-interference capabilities is provided. When using FHC in an environment, sharing time and frequency resources, avoiding collisions, and differentiating services are all made more complex as a result of this. A protocol for FHC that is based on dispersed wireless networks is presented by the authors of this research. It is a mechanism for multiple access control, which is prioritised and distributed. The ratio of empty channels metric can be found in the previous sentence. It is possible to provide priority in channel access by assigning different preset ratios of empty channel thresholds to the various traffic classes. Frames from frequency spread segments that have a partial collision are included as well. An analytical model is simulated for the analysis in terms of collision probability, transmission probability, and frame service time in order to carry out a theoretical examination of the performance of FHC. The objective of this inquiry is to determine how well FHC works. The analytical model has been proven correct by the exhaustive simulations as well as the theoretical findings. Cloud platforms are often used in the instruction of the most cutting-edge machine learning techniques of today, such as deep neural networks. This is done in order to take advantage of the cloud's capacity to scale elastically. In order to satisfy the criteria of these sorts of applications, federated learning, has been proposed as a distributed machine learning solution. This is done in order to fulfil the requirements of these kinds of applications. In federated learning (FL), even though everyone who uses the system works together to train a model, nobody ever shares their data with anybody else. Each user trains a local model with their own data, and then communicates the updated models with a FL server so that the data can be aggregated and a global model can be constructed. This process ensures that each user's model is unique. This process is repeated until a global model has been developed. This kind of training not only reduces the amount of network overhead that is necessary to transfer data to a centralised server, but it also safeguards the personal information of the users. Within the framework of this work, we looked at the feasibility of using the FL technique of learning on the many devices that are part of the dispersed network. On a centralised server, we conduct an analysis of the performance of the FL model by comparing its accuracy and the amount of time it takes to train using a range of various parameter value combinations. Additionally, the accuracy of these federated models may be made to reach a level that is comparable to that of the accuracy of central models.

ML Pro: digital assistance system for interactive machine learning in production

The application of machine learning promises great growth potential for industrial production. The development process of a machine learning solution for industrial use cases requires multi-layered, sophisticated decision-making processes along the pipeline that can only be accomplished by subject matter experts with knowledge of statistical mathematics, coding, and engineering process knowledge. By having humans and computers work together in a digital assistance system, the special characteristics of human and artificial intelligence can be used synergistically. This paper presents the development of a digital human-centered assistance system for employees in the production and development departments of industrial manufacturing companies. This assistance system enables users to apply production-specific data mining and machine learning techniques without programming to typical tabular production data, which is often inherently high-dimensional, nonstationary, and highly imbalanced data streams. Through tight interactive process guidance that considers the dependencies between machine learning process modules, users are empowered to build and optimize predictive models. Compared to existing commercial and academic tools with similar objectives, the digital assistance system offers the added value that both classical shallow and deep learning as well as generative and oversampling methods can be interactively applied to all feature table use cases for different user modes without programming.

Online Extra Trees Regressor.

Data production has followed an increased growth in the last years, to the point that traditional or batch machine-learning (ML) algorithms cannot cope with the sheer volume of generated data. Stream or online ML presents itself as a viable solution to deal with the dynamic nature of streaming data. Besides coping with the inherent challenges of streaming data, online ML solutions must be accurate, fast, and bear a reduced memory footprint. We propose a new decision tree-based ensemble algorithm for online ML regression named online extra trees (OXT). Our proposal takes inspiration from the batch learning extra trees (XT) algorithm, a popular and faster alternative to random forest (RF). While speed and memory costs might not be a central concern in most batch applications, they become crucial in data stream data learning. Our proposal combines subbagging (sampling without replacement), random tree split points, and model trees to deliver competitive prediction errors and reduced computational costs. Throughout an extensive experimental evaluation comprising 22 real-world and synthetic datasets, we compare OXT against the state-of-the-art adaptive RF (ARF) and other incremental regressors. OXT is generally more accurate than its competitors while running significantly faster than ARF and expending significantly less memory.

Accurate prediction of calving in dairy cows by applying feature engineering and machine learning.

Prediction of calving is key to dairy cow management. Current trends of increasing herd sizes globally can directly impact the time that farmers spend monitoring individual animals. Automated monitoring on behavioural and physiological changes prior to parturition can be used to develop machine learning solutions for calving prediction. In this study, we developed a machine learning algorithm for the prediction of calving in dairy cows. We demonstrated that temperature and activity index information retrieved from a commercial reticuloruminal bolus sensor can accurately predict calving from 1-day to 5-days in advance. The best prediction solution using data from 82 dairy cows, achieved up to 87.81 % in accuracy, 92.99 % in specificity, 75.84 % in sensitivity, 82.99 % in positive predictive value (PPV), 78.85 % in F-score, and 90.02 % in negative predictive value (NPV) on the test dataset when using information from 2-days in advance and all the subsets of feature characteristics (temperature + drinking + activity). The performance only decreased by 2.45 % points in accuracy, 0.74 % points in specificity, 6.41 % points in sensitivity, 2.45 % points in positive predictive value, 4.91 % points in F-score, and 2.44 % points in negative predictive value on the test dataset when using all feature characteristics and 5-days in advance information compared to using all features and information from 2-days in advance. Full evaluation of the performance of the prediction showed an improvement when using all the different subsets of feature characteristics together (temperature, activity, and drinking) compared to using temperature features only. When adding activity and drinking to the subset of temperature features, an average increase of 2.70, 1.52, 5.40, 4.39, 5.02, 2.13 % points in accuracy, specificity, sensitivity, PPV, F-score, and NPV, respectively, was obtained. Notably, evaluation of feature importance (i.e., relative weight of any given feature in relation to model prediction) showed that 3-5 (depending on the selected days in advance model) of the top ten features were derived from drinking behaviour, showing the relevance that this behaviour can have in the prediction of calving. This algorithm can provide a useful tool for automated calving prediction in dairy cows which has potential for improvement of health, welfare, and productivity in the dairy industry.

Recurrence-Aware Long-Term Cognitive Network for Explainable Pattern Classification.

Machine-learning solutions for pattern classification problems are nowadays widely deployed in society and industry. However, the lack of transparency and accountability of most accurate models often hinders their safe use. Thus, there is a clear need for developing explainable artificial intelligence mechanisms. There exist model-agnostic methods that summarize feature contributions, but their interpretability is limited to predictions made by black-box models. An open challenge is to develop models that have intrinsic interpretability and produce their own explanations, even for classes of models that are traditionally considered black boxes like (recurrent) neural networks. In this article, we propose a long-term cognitive network (LTCN) for interpretable pattern classification of structured data. Our method brings its own mechanism for providing explanations by quantifying the relevance of each feature in the decision process. For supporting the interpretability without affecting the performance, the model incorporates more flexibility through a quasi-nonlinear reasoning rule that allows controlling nonlinearity. Besides, we propose a recurrence-aware decision model that evades the issues posed by the unique fixed point while introducing a deterministic learning algorithm to compute the tunable parameters. The simulations show that our interpretable model obtains competitive results when compared to state-of-the-art white and black-box models.

AIMS: An Automatic Semantic Machine Learning Microservice Framework to Support Biomedical and Bioengineering Research.

The fusion of machine learning and biomedical research offers novel ways to understand, diagnose, and treat various health conditions. However, the complexities of biomedical data, coupled with the intricate process of developing and deploying machine learning solutions, often pose significant challenges to researchers in these fields. Our pivotal achievement in this research is the introduction of the Automatic Semantic Machine Learning Microservice (AIMS) framework. AIMS addresses these challenges by automating various stages of the machine learning pipeline, with a particular emphasis on the ontology of machine learning services tailored to the biomedical domain. This ontology encompasses everything from task representation, service modeling, and knowledge acquisition to knowledge reasoning and the establishment of a self-supervised learning policy. Our framework has been crafted to prioritize model interpretability, integrate domain knowledge effortlessly, and handle biomedical data with efficiency. Additionally, AIMS boasts a distinctive feature: it leverages self-supervised knowledge learning through reinforcement learning techniques, paired with an ontology-based policy recording schema. This enables it to autonomously generate, fine-tune, and continually adapt to machine learning models, especially when faced with new tasks and data. Our work has two standout contributions demonstrating that machine learning processes in the biomedical domain can be automated, while integrating a rich domain knowledge base and providing a way for machines to have self-learning ability, ensuring they handle new tasks effectively. To showcase AIMS in action, we have highlighted its prowess in three case studies of biomedical tasks. These examples emphasize how our framework can simplify research routines, uplift the caliber of scientific exploration, and set the stage for notable advances.

Optimal Deep-Learning-Based Cyberattack Detection in a Blockchain-Assisted IoT Environment

The Internet of Things (IoT) is the most extensively utilized technology nowadays that is simple and has the advantage of replacing the data with other devices by employing cloud or wireless networks. However, cyber-threats and cyber-attacks significantly affect smart applications on these IoT platforms. The effects of these intrusions lead to economic and physical damage. The conventional IoT security approaches are unable to handle the current security problems since the threats and attacks are continuously evolving. In this background, employing Artificial Intelligence (AI) knowledge, particularly Machine Learning (ML) and Deep Learning (DL) solutions, remains the key to delivering a dynamically improved and modern security system for next-generation IoT systems. Therefore, the current manuscript designs the Honey Badger Algorithm with an Optimal Hybrid Deep Belief Network (HBA-OHDBN) technique for cyberattack detection in a blockchain (BC)-assisted IoT environment. The purpose of the proposed HBA-OHDBN algorithm lies in its accurate recognition and classification of cyberattacks in the BC-assisted IoT platform. In the proposed HBA-OHDBN technique, feature selection using the HBA is implemented to choose an optimal set of features. For intrusion detection, the HBA-OHDBN technique applies the HDBN model. In order to adjust the hyperparameter values of the HDBN model, the Dung Beetle Optimization (DBO) algorithm is utilized. Moreover, BC technology is also applied to improve network security. The performance of the HBA-OHDBN algorithm was validated using the benchmark NSLKDD dataset. The extensive results indicate that the HBA-OHDBN model outperforms recent models, with a maximum accuracy of 99.21%.

Machine Learning Solutions for Offshore Wind Farms: A Review of Applications and Impacts

The continuous advancement within the offshore wind energy industry is propelled by the imperatives of renewable energy generation, climate change policies, and the zero-emission targets established by governments and communities. Increasing the dimensions of offshore wind turbines to augment energy production, enhancing the power generation efficiency of existing systems, mitigating the environmental impacts of these installations, venturing into deeper waters for turbine deployment in regions with optimal wind conditions, and the drive to develop floating offshore turbines stand out as significant challenges in the domains of development, installation, operation, and maintenance of these systems. This work specifically centers on providing a comprehensive review of the research undertaken to tackle several of these challenges using machine learning and artificial intelligence. These machine learning-based techniques have been effectively applied to structural health monitoring and maintenance, facilitating the more accurate identification of potential failures and enabling the implementation of precision maintenance strategies. Furthermore, machine learning has played a pivotal role in optimizing wind farm layouts, improving power production forecasting, and mitigating wake effects, thereby leading to heightened energy generation efficiency. Additionally, the integration of machine learning-driven control systems has showcased considerable potential for enhancing the operational strategies of offshore wind farms, thereby augmenting their overall performance and energy output. Climatic data prediction and environmental studies have also benefited from the predictive capabilities of machine learning, resulting in the optimization of power generation and the comprehensive assessment of environmental impacts. The scope of this review primarily includes published articles spanning from 2005 to March 2023.

Machine Learning Solution for Predicting Vibrations while Drilling the Curve Section.

The downhole vibration is one of the most crucial factors that affect downhole equipment performance and failure, besides wellbore instability. Downhole tool failure, hole problems, mechanical energy loss, and ineffective drilling performance are commonly associated with drillstring high vibration levels. The high vibration level will lead to more complications while drilling that might cause nonproductive time and extra cost. Meanwhile, the downhole sensors for detecting the drillstring vibrations add more cost to the operation. Consequently, the new solutions based on technology capabilities provide a powerful tool to integrate and interpret the drilling data for the best use of @@the data for operation performance enhancement. This study provides a successful application for utilizing the surface drilling data to automate drillstring vibration detection during the drilling curve section employing machine learning (ML) techniques. The axial, torsional, and lateral vibration modes are detected through testing four ML techniques named the @@adaptive neuro-fuzzy inference system (ANFIS), radial basis function (RBF), functional networks (FN), and support vector machines (SVMs) with real field data. The models' development was achieved by comprehensive study starting from data gathering, wrangling, statistical analysis, developing the ML models, evaluating the model prediction accuracy, and reporting the high accuracy results. The developed models were evaluated, and results showed that ANFIS and SVM models provided the highest accuracy with a coefficient of correlation (R) ranging from 0.9 to 0.99 followed by the RBF and FN models through model training and testing (R ranging from 0.82 to 0.96). Validating the models over unseen data confirmed the high accuracy prediction for the three vibration modes. Generally, the developed models provided technically accepted accuracy with R higher than 0.93 and AAPE less than 2.8% for SVM and ANFIS models while FN and RBF showed R between 0.82 and 0.95 and AAPE less than 5.7% between actual readings and predictions. Based on these results, the developed ML algorithm might be utilized as an intelligent solution to autodetect downhole vibration while drilling from surface sensor data only, which will save the downhole tool cost.

ANOMALY DETECTION IN THE TEMPERATURE OF AN AC MOTOR USING EMBEDDED MACHINE LEARNING

The integration of machine learning solutions is becoming more prominent in the industry. In industrial maintenance, new approaches categorized under predictive maintenance primarily use machine learning to identify patterns that could lead to machine failures. However, in most cases, implementing a machine learning approach is very expensive regarding resources and experienced personnel. Therefore, this approach is usually more costly in some machines than replacing these faulty machines instead. This paper proposes a low-cost machine-learning approach to detect anomalies in a rotary machine by monitoring its casing temperature using EdgeImpulse to Train the model and a Raspberry Pico as the microcontroller. The project is divided into two phases. Data is collected to be used to train and test the model. The model is then deployed to the microcontroller and is connected to a sensor attached to the motor. The model developed showed promising results with an accuracy of 91% and a ƒ1 score of 0.91.

Contemporary challenges and AI solutions in port operations: applying Gale–Shapley algorithm to find best matches

Artificial intelligence (AI) developments enable human capability to deliver the same outcome at a lower cost. This research performs a high-level matching between AI solutions and challenges within the port area by developing a novel academic approach. This way, the matching is carried out more structured than when one (manager, developer, challenge owner, etc.) fulfils it based on their opinion without following any structured approach. Therefore, the study defines first a comprehensive typology of port stakeholders' challenges, which can be solved via AI solutions. This typology presents challenges, including their main issues, widespread impact, and potential solutions. A state-of-the-art review of AI solutions that can address these challenges is carried out in parallel. Secondly, this review clearly distinguishes between AI solutions based on their technology and functionality. Thirdly, this research selects an appropriate AI solution for addressing each identified challenge in the port operation by upgrading the Gale–Shapley algorithm. Finally, it shows that the most critical presented AI solutions in this study use various machine learning (ML) techniques. Besides, concerning the AI solution's reusability feature and the result of high-level matching, this research shows that the implementation phase effort can be drastically reduced by using the recently developed matching algorithm.

A Comprehensive Review and a Taxonomy of Edge Machine Learning: Requirements, Paradigms, and Techniques

The union of Edge Computing (EC) and Artificial Intelligence (AI) has brought forward the Edge AI concept to provide intelligent solutions close to the end-user environment, for privacy preservation, low latency to real-time performance, and resource optimization. Machine Learning (ML), as the most advanced branch of AI in the past few years, has shown encouraging results and applications in the edge environment. Nevertheless, edge-powered ML solutions are more complex to realize due to the joint constraints from both edge computing and AI domains, and the corresponding solutions are expected to be efficient and adapted in technologies such as data processing, model compression, distributed inference, and advanced learning paradigms for Edge ML requirements. Despite the fact that a great deal of the attention garnered by Edge ML is gained in both the academic and industrial communities, we noticed the lack of a complete survey on existing Edge ML technologies to provide a common understanding of this concept. To tackle this, this paper aims at providing a comprehensive taxonomy and a systematic review of Edge ML techniques, focusing on the soft computing aspects of existing paradigms and techniques. We start by identifying the Edge ML requirements driven by the joint constraints. We then extensively survey more than twenty paradigms and techniques along with their representative work, covering two main parts: edge inference, and edge learning. In particular, we analyze how each technique fits into Edge ML by meeting a subset of the identified requirements. We also summarize Edge ML frameworks and open issues to shed light on future directions for Edge ML.