Evaluating classification models for resource-constrained hardware

A hybrid wired/wireless SDN switch enables unified control across mixed-access networks but faces challenges in real-time traffic classification and flow management due to the lack of intelligence. This paper presents an intelligent hybrid SDN switch augmented with embedded machine learning (ML) models for accurate traffic classification and low-latency prediction. We evaluated six supervised ML algorithms, Logistic Regression, Random Forest, Support Vector Machine, Naive Bayes, k-Nearest Neighbors, and Neural Network, under both single-flow and concurrent multiflow scenarios using real traffic traces, including DNS, Ping, Telnet, and interactive gaming protocols. The evaluation results show that the Random Forest model consistently outperforms others, achieving classification accuracies of 99.97% in single-flow and 99.86% in multi-flow scenarios, while accurately predicting the traffic in millisecond-level inference latency. These findings demonstrate the feasibility of embedding ML-driven intelligence into hybrid SDN switches to support real-time, application-aware traffic control in emerging network infrastructures.

Background Current heart failure guidelines emphasize the importance of timely detection of subclinical left ventricular (LV) remodelling and dysfunction for more precise risk stratification of asymptomatic subjects. Both LV diastolic dysfunction (LVDD) and LV hypertrophy (LVH) as assessed by echocardiography are known independent prognostic markers of future cardiovascular events in the community. However, selective screening strategies of individuals at risk who would benefit most from in-depth cardiac phenotyping are lacking. Purpose We assess the utility of several Machine Learning (ML) classifiers built on clinical and biochemical features for detecting subclinical LV abnormalities. Methods We included 1407 participants (mean age, 51 years, 51% women) randomly recruited from the general population. We used echocardiographic parameters reflecting LV diastolic function and structure to define LV abnormalities (LVDD, n=239; LVH, n=135). After that four supervised ML algorithms (Random Forest (RF), Gradient Boosting (GD), Stochastic Gradient Descent (SGD) and Support Vector Machines (SV)) were built based on routine clinical, hemodynamic and laboratory data (features; n=61) to categorize LVDD and LVH (two prediction tasks). We applied a 10-fold stratified cross-validation set-up. Results ML classifiers exhibited a high area under the ROC (AUC) for predicting LVDD with values between 88.5% and 93.1% (Figure, left panel). Age, BMI, different components of blood pressure, antihypertensive treatment, routine biomarkers such as serum electrolytes, creatinine, blood sugar, leptin, uric acid, lipid profile, as well as blood cell counts were the top selected features for predicting LVDD. Prediction AUC of ML algorithms for detection of LVH was somewhat lower than for LVDD and ranged from 72.5% to 78.7% (Figure, right panel). The top selected features for LVH classifier were similar to those of LVDD, but also included social class, serum gamma-glutamyl transferase, fasting insulin, plasma renin activity and cortisol. ROC curves (sensitivity-1-specificity) Conclusions ML algorithms combining routinely measured clinical and laboratory data have shown high accuracy of LVDD and LVH prediction. These ML classifiers might be useful to preselect individuals at risk for further in depth echocardiographic examination, monitoring and implementation of preventive strategies in order to delay transition to disease symptoms.

Objective: This study aimed to use machine learning (ML) for the prediction of intraoperative packed red cell (PRC) transfusion and the number of units of transfused PRC, as well as estimate the net benefit of the ML models through decision curve analysis. Methods: The retrospective cohort study was conducted on patients who underwent cranial operations. Clinical data and transfusion data were extracted. Supervised ML algorithms were trained and tested as ML classification for the prediction of intraoperative PRC transfusion and ML regression for predicting the number of transfused PRC units. Results: Out of 2683 patients, 42.9% of neurosurgical patients intraoperatively received PRC. Artificial neural network, gradient boosting classifier, and random forest were the algorithms that had high area under the receiver operating characteristic curve of 0.912, 0.911, and 0.909, respectively, in ML classification, while random forest with regression had the lowest root mean squared error and mean absolute error in ML regression. Conclusions: ML is one of the most effective approaches to developing clinical prediction tools that can enhance the efficiency of blood utilization. Additionally, ML has become a valuable tool in modern health technologies as the computerized clinical decision support systems assist the physician in decision-making in real-world practice.

Surgical site infection (SSI) following a neurosurgical operation is a complication that impacts morbidity, mortality, and economics. Currently, machine learning (ML) algorithms are used for outcome prediction in various neurosurgical aspects. The implementation of ML algorithms to learn from medical data may help in obtaining prognostic information on diseases, especially SSIs. The purpose of this study was to compare the performance of various ML models for predicting surgical infection after neurosurgical operations. A retrospective cohort study was conducted on patients who had undergone neurosurgical operations at tertiary care hospitals between 2010 and 2017. Supervised ML algorithms, which included decision tree, naive Bayes with Laplace correction, k-nearest neighbors, and artificial neural networks, were trained and tested as binary classifiers (infection or no infection). To evaluate the ML models from the testing data set, their sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), as well as their accuracy, receiver operating characteristic curve, and area under the receiver operating characteristic curve (AUC) were analyzed. Data were available for 1471 patients in the study period. The SSI rate was 4.6%, and the type of SSI was superficial, deep, and organ/space in 1.2%, 0.8%, and 2.6% of cases, respectively. Using the backward stepwise method, the authors determined that the significant predictors of SSI in the multivariable Cox regression analysis were postoperative CSF leakage/subgaleal collection (HR 4.24, p < 0.001) and postoperative fever (HR 1.67, p = 0.04). Compared with other ML algorithms, the naive Bayes had the highest performance with sensitivity at 63%, specificity at 87%, PPV at 29%, NPV at 96%, and AUC at 76%. The naive Bayes algorithm is highlighted as an accurate ML method for predicting SSI after neurosurgical operations because of its reasonable accuracy. Thus, it can be used to effectively predict SSI in individual neurosurgical patients. Therefore, close monitoring and allocation of treatment strategies can be informed by ML predictions in general practice.

In Mild Cognitive Impairment (MCI), identifying a high risk of conversion to Alzheimer’s Disease Dementia (AD) is a primary goal for patient management. Machine Learning (ML) algorithms are widely employed to pursue data-driven diagnostic and prognostic goals. An agreement on the stability of these algorithms –when applied to different biomarkers and other conditions– is far from being reached. In this study, we compared the different prognostic performances of three supervised ML algorithms fed with multimodal biomarkers of MCI subjects obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. Random Forest, Gradient Boosting, and eXtreme Gradient Boosting algorithms predict MCI conversion to AD. They can also be simultaneously employed –with the voting procedure– to improve predictivity. AD prediction accuracy is influenced by the nature of the data (i.e., neuropsychological test scores, cerebrospinal fluid AD-related proteins and APOE ε4, cerebral structural MRI (sMRI) data). In our study, independent of the applied ML algorithms, sMRI data showed the lowest accuracy (0.79) compared to other classes. Multimodal data were helpful in the algorithms’ performances by combining clinical and biological measures. Accordingly, using the three ML algorithms, the highest accuracy (0.90) was reached by employing neuropsychological and AD-related biomarkers. Finally, the feature selection procedure indicated that the most critical variables in the respective classes were the ADAS-Cog-13 scale, the medial temporal lobe and hippocampus atrophy, and the ratio between phosphorylated Tau and Aβ42 proteins. In conclusion, our data support the notion that using multiple ML algorithms and multimodal biomarkers helps make more accurate and solid predictions.

Background: Currently, Parkinson’s disease (PD) diagnosis is mainly based on medical history and physical examination, and there is no objective and consistent basis. By the time of diagnosis, the disease would have progressed to the middle and late stages. Pilot studies have shown that a unique smell was present in the skin sebum of PD patients. This increases the possibility of a noninvasive diagnosis of PD using an odor profile. Methods: Fast gas chromatography (GC) combined with a surface acoustic wave sensor with embedded machine learning (ML) algorithms was proposed to establish an artificial intelligent olfactory (AIO) system for the diagnosis of Parkinson’s through smell. Sebum samples of 43 PD patients and 44 healthy controls (HCs) from Fourth Affiliated Hospital of Zhejiang University School of Medicine, China, were smelled by the AIO system. Univariate and multivariate methods were used to identify the significant volatile organic compound (VOC) features in the chromatograms. ML algorithms, including support vector machine, random forest (RF), k nearest neighbor (KNN), AdaBoost (AB), and Naive Bayes (NB), were used to distinguish PD patients from HC based on the VOC peaks in the chromatograms of sebum samples. Results: VOC peaks with average retention times of 5.7, 6.0, and 10.6 s, respectively, corresponding to octanal, hexyl acetate, and perillic aldehyde, were significantly different in PD and HC. The accuracy of the classification based on the significant features was 70.8%. Based on the odor profile, the classification had the highest accuracy and F1 of the five models with 0.855 from NB and 0.846 from AB, respectively, in the process of model establishing. The highest specificity and sensitivity of the five classifiers were 91.6% from NB and 91.7% from RF and KNN, respectively, in the evaluating set. Conclusions: The proposed AIO system can be used to diagnose PD through the odor profile of sebum. Using the AIO system is helpful for the screening and diagnosis of PD and is conducive to further tracking and frequent monitoring of the PD treatment process.

Feature extraction-reduction and machine learning for fault diagnosis in PV panels

Machine learning is an expanding field with an ever-increasing role in everyday life, with its utility in the industrial, agricultural, and medical sectors being undeniable. Recently, this utility has come in the form of machine learning implementation on embedded system devices. While there have been steady advances in the performance, memory, and power consumption of embedded devices, most machine learning algorithms still have a very high power consumption and computational demand, making the implementation of embedded machine learning somewhat difficult. However, different devices can be implemented for different applications based on their overall processing power and performance. This paper presents an overview of several different implementations of machine learning on embedded systems divided by their specific device, application, specific machine learning algorithm, and sensors. We will mainly focus on NVIDIA Jetson and Raspberry Pi devices with a few different less utilized embedded computers, as well as which of these devices were more commonly used for specific applications in different fields. We will also briefly analyze the specific ML models most commonly implemented on the devices and the specific sensors that were used to gather input from the field. All of the papers included in this review were selected using Google Scholar and published papers in the IEEExplore database. The selection criterion for these papers was the usage of embedded computing systems in either a theoretical study or practical implementation of machine learning models. The papers needed to have provided either one or, preferably, all of the following results in their studies—the overall accuracy of the models on the system, the overall power consumption of the embedded machine learning system, and the inference time of their models on the embedded system. Embedded machine learning is experiencing an explosion in both scale and scope, both due to advances in system performance and machine learning models, as well as greater affordability and accessibility of both. Improvements are noted in quality, power usage, and effectiveness.

In recent years, the microbiome has been recognized as an important factor associated with cardiovascular disease (CVD), which is the leading cause of human mortality worldwide. Disparities in gut microbial compositions between individuals with and without CVD were reported, whereby, we hypothesized that utilizing such microbiome-based data for training with supervised machine learning (ML) models could be exploited as a new strategy for evaluation of cardiovascular health. To test our hypothesis, we analyzed the metagenomics data extracted from the American Gut Project. Specifically, 16S rRNA reads from stool samples of 478 CVD and 473 non-CVD control samples were analyzed using five supervised ML algorithms: random forest (RF), support vector machine with radial kernel (svmRadial), decision tree (DT), elastic net (ENet) and neural networks (NN). Thirty-nine differential bacterial taxa (LEfSe: LDA &gt; 2) were identified between CVD and non-CVD groups. ML classifications, using these taxonomic features, achieved an AUC (area under the receiver operating characteristic curve) of ~0.58 (RF). However, by choosing the top 500 high-variance features of operational taxonomic units (OTUs) for training ML models, an improved AUC of ~0.65 (RF) was achieved. Further, by limiting the selection to only the top 25 highly contributing OTU features to reduce the dimensionality of feature space, the AUC was further significantly enhanced to ~0.70 (RF). In summary, this study is the first to demonstrate the successful development of a ML model using microbiome-based datasets for a systematic diagnostic screening of CVD.

Three-phase induction motors (IMs) are one of the most employed electric machines in industrial and household applications. Condition monitoring of these machines is essential to avoid unplanned maintenance and thereby enhance the availability. Artificial Intelligence (AI) technologies are emerging as an advanced tool for automating condition monitoring process to detect incipient faults at early stages. Machine Learning (ML) algorithms have been identified as a promising approach for condition monitoring of IMs and predicting maintenance to avoid failures. However, selecting the suitable ML algorithm for a given application is challenging because there is no predefined set of application-based algorithms. In addition, raw data processing and feature selection need careful attention to improve the accuracy of the results. This paper reviews supervised ML algorithms that can be used for condition monitoring of IMs and identifies their benefits and drawbacks. It then discusses how the dominant features from raw data can be selected through time domain and frequency domain analysis using the acoustic data collected from a three-phase induction motor. The study investigates classification accuracy of each ML algorithm and a procedure for selecting an algorithm based on the experimental results. Results of this study show that Support Vector Machines (SVM) algorithm outperforms other competing algorithms in condition monitoring of IMs when the dominant frequency components obtained through Fast Fourier Transform (FFT) are used as training data.

An intrusion detection system (IDS) is an important protection instrument for detecting complex network attacks. Various machine learning (ML) or deep learning (DL) algorithms have been proposed for implementing anomaly-based IDS (AIDS). Our review of the AIDS literature identifies some issues in related work, including the randomness of the selected algorithms, parameters, and testing criteria, the application of old datasets, or shallow analyses and validation of the results. This paper comprehensively reviews previous studies on AIDS by using a set of criteria with different datasets and types of attacks to set benchmarking outcomes that can reveal the suitable AIDS algorithms, parameters, and testing criteria. Specifically, this paper applies 10 popular supervised and unsupervised ML algorithms for identifying effective and efficient ML-AIDS of networks and computers. These supervised ML algorithms include the artificial neural network (ANN), decision tree (DT), k-nearest neighbor (k-NN), naive Bayes (NB), random forest (RF), support vector machine (SVM), and convolutional neural network (CNN) algorithms, whereas the unsupervised ML algorithms include the expectation-maximization (EM), k-means, and self-organizing maps (SOM) algorithms. Several models of these algorithms are introduced, and the turning and training parameters of each algorithm are examined to achieve an optimal classifier evaluation. Unlike previous studies, this study evaluates the performance of AIDS by measuring the true positive and negative rates, accuracy, precision, recall, and F-Score of 31 ML-AIDS models. The training and testing time for ML-AIDS models are also considered in measuring their performance efficiency given that time complexity is an important factor in AIDSs. The ML-AIDS models are tested by using a recent and highly unbalanced multiclass CICIDS2017 dataset that involves real-world network attacks. In general, the k-NN-AIDS, DT-AIDS, and NB-AIDS models obtain the best results and show a greater capability in detecting web attacks compared with other models that demonstrate irregular and inferior results.

Despite the promise for performance and accuracy improvements of quantum inspired (QI) algorithms over classical machine learning (ML) algorithms, such gains have not been realized in practice. The quantum inspired algorithms can theoretically achieve significant speed up based on sampling assumptions and have thus far failed to outperform the existing classical ML models in practical applications. The speedup of quantum machine learning (QML) algorithms assume the access to data in quantum random access memory (QRAM) which is a strong assumption with current quantum architectures. QI algorithms assume sample and query (SQ) access to input vector and norms of matrices using a dynamic data structure. We explore the components of these models and the assumptions in this paper by surveying the recent works in QML and QI Machine learning (QIML) algorithms. We limit our study to QML and QIML models on achieving a speed up over classical ML techniques rather than individual proofs of these algorithms. This study highlights the assumptions being made that are currently not practical for QML and QIML algorithms in achieving performance advantage over classical ML algorithms.

Embedded systems technology is undergoing a phase of transformation owing to the novel advancements in computer architecture and the breakthroughs in machine learning applications. The areas of applications of embedded machine learning (EML) include accurate computer vision schemes, reliable speech recognition, innovative healthcare, robotics, and more. However, there exists a critical drawback in the efficient implementation of ML algorithms targeting embedded applications. Machine learning algorithms are generally computationally and memory intensive, making them unsuitable for resource-constrained environments such as embedded and mobile devices. In order to efficiently implement these compute and memory-intensive algorithms within the embedded and mobile computing space, innovative optimization techniques are required at the algorithm and hardware levels. To this end, this survey aims at exploring current research trends within this circumference. First, we present a brief overview of compute intensive machine learning algorithms such as hidden Markov models (HMM), k-nearest neighbors (k-NNs), support vector machines (SVMs), Gaussian mixture models (GMMs), and deep neural networks (DNNs). Furthermore, we consider different optimization techniques currently adopted to squeeze these computational and memory-intensive algorithms within resource-limited embedded and mobile environments. Additionally, we discuss the implementation of these algorithms in microcontroller units, mobile devices, and hardware accelerators. Conclusively, we give a comprehensive overview of key application areas of EML technology, point out key research directions and highlight key take-away lessons for future research exploration in the embedded machine learning domain.

The look for methodologies that can make inferences from externally supplied data develop broad hypotheses that are subsequently used to create forecasts concerning future events is known as supervised machine learning (SML). This study examine machine learning (ML) classification strategies, compares supervised learning algorithms, and determines foremost efficient classification algorithm based on the data set, number of instances, and variables (features). ML with the Waikato Environment for Knowledge Analysis (WEKA) tool, 7 different machine learning algorithms were considered: Decision Table, Random Forest (RF), Naive Bayes (NB), support vector machine (SVM), neural networks (Perceptron), JRip, and Decision Tree (J48). Time it takes to make a design and be concise (accuracy) are factors on the one end, and the kappa statistic and Mean Absolute Error (MAE) are factors on the other. For supervised predictive machine learning to work, machine learning algorithms must be accurate, and error-free.KeywordsMachine learningMean absolute errorWEKADeep learning

The application of machine-learning and Raman spectroscopy for the rapid detection of edible oils type and adulteration