Running Machine Learning Algorithms Research Articles

Heart Disease Prediction Using ML

Machine Learning and artificial intelligence have found valuable on variety of disciplines during their growth, particularly in the light of massive increase in data in recent years. It has the potential to be more dependable in terms of producing quicker and more accurate illness prediction judgments. Therefore, the use of machine learning algorithms to forecast different diseases is growing. Building a model can also aid in the visualization and analysis of diseases to increase the accuracy and consistency of reporting. This article has looked into using several machine learning algorithms to identify cardiac disease. This article's study has demonstrated a step procedure. In a dataset on heart disease initially prepared in the format needed to run machine learning algorithms. The UCI is the source of patient medical records and other data. The presence are absence of heart disease in patients is then ascertained using the heart disease dataset. Second, this paper presents a number of noteworthy findings. The confusion matrix is used to validate the accuracy rate of machine learning methods, including Gradient Boosting Classifier, Support Vector Machine, and Logistic Regression. According to recent research, the Logistic Regression method outperforms other algorithms in terms of accuracy, yielding a high 95% rate. It also outperforms the other four algorithms in terms of recall, precision, and f1-score correctness. The difficult and future research component of this project will be raising the accuracy rates of the machine learning algorithms to between 97% and 100%.

Geochemistry π: Automated Machine Learning Python Framework for Tabular Data

AbstractAlthough machine learning (ML) has brought new insights into geochemistry research, its implementation is laborious and time‐consuming. Here, we announce Geochemistry π, an open‐source automated ML Python framework. Geochemists only need to provide tabulated data and select the desired options to clean data and run ML algorithms. The process operates in a question‐and‐answer format, and thus does not require that users have coding experience. After either automatic or manual parameter tuning, the automated Python framework provides users with performance and prediction results for the trained ML model. Based on the scikit‐learn library, Geochemistry π has established a customized automated process for implementing classification, regression, dimensionality reduction, and clustering algorithms. The Python framework enables extensibility and portability by constructing a hierarchical pipeline architecture that separates data transmission from the algorithm application. The AutoML module is constructed using the Cost‐Frugal Optimization and Blended Search Strategy hyperparameter search methods from the A Fast and Lightweight AutoML Library, and the model parameter optimization process is accelerated by the Ray distributed computing framework. The MLflow library is integrated into ML lifecycle management, which allows users to compare multiple trained models at different scales and manage the data and diagrams generated. In addition, the front‐end and back‐end frameworks are separated to build the web portal, which demonstrates the ML model and data science workflow through a user‐friendly web interface. In summary, Geochemistry π provides a Python framework for users and developers to accelerate their data mining efficiency with both online and offline operation options.

Dirençli Punta Kaynağı Prosesinin KNN ve CART Makine Öğrenimi Teknikleri ile Değerlendirilmesi

Spot welding, a type of resistance welding, is a welding application widely used in the production area and it is a common method for joining metal sheets. The spot-welding process is widely used in many production areas, especially in the automotive industry, radiator, and wire mesh production. Spot welding in car production lines is mainly performed by robotic applications. Industry 4.0 and digital transformation trends have led to unprecedented data growth. Nowadays, the manufacturing industry benefits from the power of machine learning and data science algorithms to monitor production processes and make predictions for quality, maintenance, and production optimization. Applying machine learning algorithms reduces the duration and cost of experiments. This study aims to confirm whether the spot welding, applied by robotic arms, is within the ideal spot-welding norms, in real production area. The ideal parameter norms were evaluated by using KNN and CART machine learning algorithms. To use real production data, this study was executed in the body production assembly line, which is selected as the pilot area, at TOFAŞ factory. The data set used in this research consists of the welding parameters of the current year, 2023. By running machine learning algorithms on the dataset, the performance evaluation of each algorithm was examined and the most appropriate estimation method was determined. In the experiments, the best F1-Score value was obtained by the CART model with 93%.

PePTM: An Efficient and Accurate Personalized P2P Learning Algorithm for Home Thermal Modeling

Nowadays, the integration of home automation systems with smart thermostats is a common trend, designed to enhance resident comfort and conserve energy. The introduction of smart thermostats that can run machine learning algorithms has opened the door for on-device training, enabling customized thermal experiences in homes. However, leveraging the flexibility offered by on-device learning has been hindered by the absence of a tailored learning scheme that allows for accurate on-device training of thermal models. Traditional centralized learning (CL) and federated learning (FL) schemes rely on a central server that controls the learning experience, compromising the home’s privacy and requiring significant energy to operate. To address these challenges, we propose PePTM, a personalized peer-to-peer thermal modeling algorithm that generates tailored thermal models for each home, offering a controlled learning experience with a minimal training energy footprint while preserving the home’s privacy, an aspect difficult to achieve in both CL and FL. PePTM consists of local and collaborative learning phases that enable each home to train its thermal model and collaboratively improve it with a set of similar homes in a peer-to-peer fashion. To showcase the effectiveness of PePTM, we use a year’s worth of data from US homes to train thermal models using the RNN time-series model and compare the data across three learning schemes: CL, FL, and PePTM, in terms of model performance and the training energy footprint. Our experimental results show that PePTM is significantly energy-efficient, requiring 695 and 40 times less training energy than CL and FL, respectively, while maintaining comparable performance. We believe that PePTM sets the stage for new avenues for on-device thermal model training, providing a personalized thermal experience with reduced energy consumption and enhanced privacy.

Constrained IoT-Based Machine Learning for Accurate Glycemia Forecasting in Type 1 Diabetes Patients.

Individuals with diabetes mellitus type 1 (DM1) tend to check their blood sugar levels multiple times daily and utilize this information to predict their future glycemic levels. Based on these predictions, patients decide on the best approach to regulate their glucose levels with considerations such as insulin dosage and other related factors. Nevertheless, modern developments in Internet of Things (IoT) technology and innovative biomedical sensors have enabled the constant gathering of glucose level data using continuous glucose monitoring (CGM) in addition to other biomedical signals. With the use of machine learning (ML) algorithms, glycemic level patterns can be modeled, enabling accurate forecasting of this variable. Constrained devices have limited computational power, making it challenging to run complex machine learning algorithms directly on these devices. However, by leveraging edge computing, using lightweight machine learning algorithms, and performing preprocessing and feature extraction, it is possible to run machine learning algorithms on constrained devices despite these limitations. In this paper we test the burdens of some constrained IoT devices, probing that it is feasible to locally predict glycemia using a smartphone, up to 45 min in advance and with acceptable accuracy using random forest.

Identifying Chinese social media users' need for affect from their online behaviors.

The need for affect (NFA), which refers to the motivation to approach or avoid emotion-inducing situations, is a valuable indicator of mental health monitoring and intervention, as well as many other applications. Traditionally, NFA has been measured using self-reports, which is not applicable in today's online scenarios due to its shortcomings in fast, large-scale assessments. This study proposed an automatic and non-invasive method for recognizing NFA based on social media behavioral data. The NFA questionnaire scores of 934 participants and their social media data were acquired. Then we run machine learning algorithms to train predictive models, which can be used to automatically identify NFA degrees of online users. The results showed that Extreme Gradient Boosting (XGB) performed best among several algorithms. The Pearson correlation coefficients between predicted scores and NFA questionnaire scores achieved 0.25 (NFA avoidance), 0.31 (NFA approach) and 0.34 (NFA total), and the split-half reliabilities were 0.66-0.70. Our research demonstrated that adolescents' NFA can be identified based on their social media behaviors, and opened a novel way of non-intrusively perceiving users' NFA which can be used for mental health monitoring and other situations that require large-scale NFA measurements.

MACHINE LEARNING AT THE EDGE: A DATA-DRIVEN ARCHITECTURE WITH APPLICATIONS TO 5G CELLULAR NETWORKS

To meet the ultra-low latency demands of future applications, the fifth generation of cellular networks (5G) will rely on edge cloud installations. In this research, we show that similar deployments may also be deployed in mobile networks to allow sophisticated data-driven and Machine Learning (ML) applications. We propose an edge- controller-based cellular network design and assess its performance using real-world data from hundreds of base stations of a large US operator. In this context, we will discuss how to dynamically cluster and associate base stations and controllers based on users' worldwide movement patterns. The controllers will then be used to run ML algorithms to forecast the number of users in each base station, as well as a use case in which these predictions are utilized by a higher-layer application to direct vehicular traffic based on network Key Performance Indicators (KPIs). We demonstrate that prediction accuracy increases when based on machine learning algorithms that depend on the controllers' view and, as a result, on the spatial correlation provided by user mobility, compared to when the prediction is based only on the local data of each individual base station.

System for Semi-Automated Literature Review Based on Machine Learning

This paper presents the design and implementation of a system for semi-automating the literature review process based on machine learning. By using machine learning algorithms, the system determines whether scientific papers belong to the topic that is being explored as part of the review process. The system’s user interface allows the process of creating a literature review to be managed through a series of steps: selecting data sources, building queries and topic searches, displaying the scientific papers found, selecting papers that belong to the set of desired papers, running machine learning algorithms for learning and automated classification, and displaying and exporting the final set of papers. Manual literature reviews are compared with automated reviews, and similarities and differences between the two approaches in terms of duration, accuracy, and ease of use are discussed. This study concludes that the best results in terms of sensitivity and accuracy for the automated literature review process are achieved by using a combined machine learning model, which uses multiple unweighted machine learning models. Cross-testing the models on two alternative datasets revealed an overlap in the machine learning hyperparameters. The stable sensitivity and accuracy in the tests indicate the potential for generalized use in automated literature review.

In preprints: improving and interrogating embryo models.

How the complexity of an embryo arises is a question at the core of developmental biology. Key discoveries have been made by observing how embryos develop. In a complementary way, a recent bottom-up approach has harnessed the ability of stem cells to self-organize into structures resembling embryos at various stages. These embryo models go under different names according to the stages or tissues they mimic (e.g. blastoids, gastruloids, axioloids, trunk-like structures, etc.) and, when efficient, predictable and predictive of development, they expand our ability to investigate the path to embryo formation by overcoming technical and ethical issues associated with embryos. Here, we highlight two recent preprints that make important steps towards methods to improve and interrogate embryo models. The first preprint (Anand et al., 2022 preprint) tackles the problem of efficiency and reproducibility in an original way by smartly combining engineering with machine learning and, subsequently, interrogating the properties underlying robustness. The second preprint (Wehmeyer et al., 2022 preprint; now published in Development) cleverly shows how chimeric gastruloids built using fluorescently labelled cells carrying different genotypes produce new insights into well-hidden functions of classical developmental genes. Together, these two papers build a path toward making embryo models useful for discoveries.In their preprint, Anand et al. propose a new approach to increase the efficiency and reproducibility of embryo models through a combination of engineering and machine learning. They build a human embryo model that mimics aspects of axial elongation by growing and differentiating pluripotent stem cells on coverslips previously imprinted with randomly positioned micropatterned islands of extracellular matrix. These micropatterns allow cells to be confined in colonies of specific sizes and geometries. The authors subsequently evaluate the influence of colony density on spontaneous axial elongation and polarization. Polarization is quantified by measuring the asymmetry of colonies by staining for CDX2, a marker of axial progenitors termed neuromesodermal progenitors (NMPs). As noticed previously (Blin et al., 2018), there is a certain variability in the polarization of colonies that is thought to originate from suboptimal boundary conditions. However, Anand et al. note that there is less variability between embryo models that occupy the same position in coverslips with identically distributed micropatterns. This suggests that the embryo models are somewhat coupled and can influence each other's ability to elongate. Accordingly, it has been reported previously that the geometric arrangement of multiple colonies affects their ability to pattern spontaneously (Blin et al., 2018). Anand et al. thus devise a strategy to leverage these observations in order to maximize the ability of colonies to elongate. To this aim, they collect data from thousands of embryo models and run machine learning algorithms to predict their polarization given the local geometric and spatial arrangements of micropatterns. This allows them, using computer simulations, to identify the arrangement that maximizes polarization, namely six embryo models arranged in hexagonal patterns spaced by 1.6 mm.In the next step, they make use of their beautifully optimized model to look for mechanisms underlying axial elongation. Through single-cell RNA sequencing and subsequent validation, they observe patterned expression of WNT5A, PRICKLE1 and FGF8 in the posterior region. By treating cells with IWP3, which inhibits porcupine O-acyltransferase, and hence WNT secretion, either at the beginning or halfway through the experiment, they see that WNT secretion is not necessary for the initial specification of CDX2+ cells, but is continuously important for the elongation process. Live imaging suggests that elongation occurs by WNT-dependent directed cell movements along the axis that are not due to a directionality of the planes of cell divisions. Through genetic engineering, they further find that the expression of CDX2, which is known to control WNT ligand expression (Chawengsaksophak et al., 2004), is necessary for elongation. Because elongation is also known to rely on FGF ligands, they test the impact of a MEK inhibitor (PD0325901) and show that this also prevents elongation. Interestingly, a pulse of FGF2 is sufficient to sustain elongation even in the absence of the canonical WNT activator CHIR99021 as it helps maintain NMPs, a phenomenon that correlates with the expression of TBXT. Given that TBXT is known to drive WNT and FGF ligand production, a feed-forward loop is likely to occur wherein WNT and FGF expressed by NMPs induce the expression of more ligands, thus sustaining tissue elongation. Hence, the authors propose that the system is excitable and entrains its own elongation. However, such excitable systems are known to be susceptible to noise and thus prone to errors. How can elongation happen in such a stereotypic way without making mistakes? Why does this stereotypic behaviour occur only when embryo models are placed at a certain distance from one another? It turns out that inhibitors of the same WNT pathway, namely SFRP1 and SFRP2, are secreted in a gradient anti-correlating the WNT one. Interestingly, when SFRP1 and SFRP2 are genetically ablated, the embryo model forms multiple NMP domains instead of one. This suggests that SFRP1 and SFRP2 are redundantly acting WNT inhibitors expressed away from the NMPs to guide the inherently fluctuating excitable systems and prevent the appearance of multiple domains. This approach is innovative, systematic, and proves useful for understanding how robustness originates from precise geometric constraints channelling intrinsic self-organized loops that are inherently prone to noise disturbances (Murray, 2013).Embryo models also lend themselves to more complex and potentially more informative manipulations. For instance, in vitro models can be formed with a mix of cells with different engineered genotypes, an operation that remains complicated with embryos. Such mixes can be useful, for example for assessing the role of intrinsic genetic programmes (e.g. Hox collinearity; Beccari et al., 2018), cellular signalling interactions (e.g. tissue inductions; Vrij et al., 2022) or the mechanisms underpinning cell–cell competition (e.g. NFKB; Lima et al., 2021). Although this idea has already been applied to build chimeric embryos, in vitro models offer a more tractable system in which cellular composition can be finely tuned and single cells can be imaged and tracked more easily.In a proof-of-principle application, Wehmeyer et al. (2022 preprint) show how chimeric gastruloids can be used to disentangle the cell-autonomous from non-cell-autonomous functions of classical gastrulation genes. They build chimeric gastruloids composed of wild-type (WT) cells mixed with cells carrying loss-of-function or inducible alleles for Eomes and brachyury (T), which encode two transcription factors that are crucial for cell differentiation during gastrulation. Thanks to fluorescent labels, cells are tracked to reveal the influence of genotype on their position, their fate and, more importantly, their impact on neighbours. For example, Wehmeyer et al. mix WT cells with cells harbouring a doxycycline-inducible expression cassette for a GFP-tagged Eomes. Forced expression of Eomes in a subset of cells cell-autonomously instructs domains of beating cardiomyocytes in ∼30% of chimeric gastruloids, which are very rarely present in gastruloids built only from WT cells. Mixing genotypes can thus add signalling or functional domains to embryo models in order to test their impact (Xu et al., 2021).In another experiment, Wehmeyer et al. generate chimeric gastruloids by mixing WT and T−/− cells. Similar to embryos, gastruloids fully depleted of T cannot elongate. However, by mixing 10% WT and 90% T−/− cells, the authors observe that T not only remains off in the knockout cells, but also ceases expression prematurely in the WT cells. This suggests a non-cell-autonomous effect of T, in agreement with previous studies showing that Nodal expression (which induces brachyury expression) is first maintained by its own expression mediated by an asymmetric enhancer and is secondly enhanced by a cascade of BMP4 and activating Wnt3a, mediated by a proximal epiblast enhancer (Plouhinec et al., 2022). To investigate the cell-autonomous effects of T expression, Wehmeyer et al. decrease the proportion of T−/− cells to 10% and analyse the spatial distribution of the mutant cells. They observe that T−/− cells in the resulting chimeric gastruloids have specific location and fate biases: they are mainly located in the mid and posterior region of the gastruloids, where they tend to differentiate towards a definitive endoderm or a neuroectodermal fate, respectively. This shows how a specific genotype can direct cells into specific fates and suggests that the gene product of T can counteract these lineage-commitment programmes. Overall, this preprint shows how chimeric gastruloids are versatile tools that can generate embryo models with an increased repertoire of cell types and be used to probe the functions of genes during development at different levels.The refinement of embryo models makes them more manipulatable, reproducible, efficient, and thus useful. How can embryo models best be used to study development? What level of fidelity is required? What can we learn from models that escape the route shaped through evolution? In our opinion, the two preprints presented here introduce experimental and computational approaches that will shape the predictive levels of embryo models and their capacity to answer new questions. As Anand et al. (2022 preprint) show, machine learning offers an efficient and targeted way to optimize protocols, while providing information on the interplay between geometric constraints and excitable molecular loops that affect axial elongation. Such information could, in turn, be tested in physical models the predictions of which can be quantitatively validated using mixed genotypes, as in chimeric gastruloids (Wehmeyer et al., 2022 preprint). Moving forward, this will help reveal the principles underlying the organization of complex tissues, organs and embryos.Wehmeyer et al., 2022 has now been published as: Wehmeyer, A. E., Schüle, K. M., Conrad, A., Schröder, C. M., Probst, S. and Arnold, S. J. (2022). Chimeric 3D gastruloids – a versatile tool for studies of mammalian peri-gastrulation development. Development 149, dev200812. doi:10.1242/dev.200812A.S. and N.R. acknowledge helpful discussions with their lab members.

A Novel Plug-in Board for Remote Insect Monitoring

The conventional approach to monitoring insect swarming is based on traps that are periodically checked by human operators. However, human checking of trap contents is expensive, and in many environments, the pest species most frequently encountered in the traps can be detected and monitored automatically. To achieve this goal, a dedicated data acquisition device is necessary, which makes real-time and online pest monitoring possible from a distant location. In addition, it is beneficial for the device to run machine learning algorithms that count and identify insects automatically from pictures. Thanks to the advantages of integrated circuits, more systems have been designed to improve integrated pest management in the context of precision agriculture. However, in our opinion, all of those systems have one or more disadvantages, such as high cost, low power autonomy, low picture quality, a WIFI coverage requirement, intensive human control, and poor software support. Therefore, the aim of this work is to present a novel plug-in board for automatic pest detection and counting. The plug-in board is dedicated to Raspberry Pi devices, especially the Raspberry Pi Zero. The proposed board, in combination with a Raspberry Pi device and a Pi camera, overcomes the limitations of other prototypes found in the literature. In this paper, a detailed description can be found about the schematic and characteristics of the board with a deep-learning-based insect-counting method.

Secure human action recognition by encrypted neural network inference

Advanced computer vision technology can provide near real-time home monitoring to support “aging in place” by detecting falls and symptoms related to seizures and stroke. Affordable webcams, together with cloud computing services (to run machine learning algorithms), can potentially bring significant social benefits. However, it has not been deployed in practice because of privacy concerns. In this paper, we propose a strategy that uses homomorphic encryption to resolve this dilemma, which guarantees information confidentiality while retaining action detection. Our protocol for secure inference can distinguish falls from activities of daily living with 86.21% sensitivity and 99.14% specificity, with an average inference latency of 1.2 seconds and 2.4 seconds on real-world test datasets using small and large neural nets, respectively. We show that our method enables a 613x speedup over the latency-optimized LoLa and achieves an average of 3.1x throughput increase in secure inference compared to the throughput-optimized nGraph-HE2.

Optimal Sampling Gaps for Adaptive Submodular Maximization

Running machine learning algorithms on large and rapidly growing volumes of data is often computationally expensive, one common trick to reduce the size of a data set, and thus reduce the computational cost of machine learning algorithms, is probability sampling. It creates a sampled data set by including each data point from the original data set with a known probability. Although the benefit of running machine learning algorithms on the reduced data set is obvious, one major concern is that the performance of the solution obtained from samples might be much worse than that of the optimal solution when using the full data set. In this paper, we examine the performance loss caused by probability sampling in the context of adaptive submodular maximization. We consider a simple probability sampling method which selects each data point with probability at least r. If we set r=1, our problem reduces to finding a solution based on the original full data set. We define sampling gap as the largest ratio between the optimal solution obtained from the full data set and the optimal solution obtained from the samples, over independence systems. Our main contribution is to show that if the sampling probability of each data point is at least r and the utility function is policywise submodular, then the sampling gap is both upper bounded and lower bounded by 1/r. We show that the property of policywise submodular can be found in a wide range of real-world applications, including pool-based active learning and adaptive viral marketing.

ArcUHI: A GIS add-in for automated modelling of the Urban Heat Island effect through machine learning

Increased urbanisation is boosting the intensity and frequency of the Urban Heat Island (UHI) effect in highly developed cities. The advances in satellite measurement are facilitating the analysis of this phenomenon using Land Surface Temperature (LST) as an indicator of the Surface UHI (SUHI). Due to the spatial implications of using remote sensing data, this research developed ArcUHI, a Geographic Information System (GIS) add-in for modelling SUHI. The tool was designed in ArcGIS, which was bound with R to run machine learning algorithms in the background. ArcUHI was tested using the metropolitan area of Madrid (Spain) in 2006, 2012 and 2018 as a case study. The add-in was found to predict observed LST with high accuracy, especially when using Random Forest Regression (RFR). Building height and albedo were identified as the main drivers of SUHI, whose magnitude and extension increased with time. In view of these results, strategic roof and wall greening was suggested as a measure to mitigate the street canyon effect entailed by buildings and offset the heat retention capacity of built-up surfaces.

Using machine learning to identify extragalactic globular cluster candidates from ground-based photometric surveys of M87

ABSTRACT Globular clusters (GCs) have been at the heart of many longstanding questions in many sub-fields of astronomy and, as such, systematic identification of GCs in external galaxies has immense impacts. In this study, we take advantage of M87’s well-studied GC system to implement supervised machine learning (ML) classification algorithms – specifically random forest and neural networks – to identify GCs from foreground stars and background galaxies, using ground-based photometry from the Canada–France–Hawaii Telescope (CFHT). We compare these two ML classification methods to studies of ‘human-selected’ GCs and find that the best-performing random forest model can reselect 61.2 per cent ± 8.0 per cent of GCs selected from HST data (ACSVCS) and the best-performing neural network model reselects 95.0 per cent ± 3.4 per cent. When compared to human-classified GCs and contaminants selected from CFHT data – independent of our training data – the best-performing random forest model can correctly classify 91.0 per cent ± 1.2 per cent and the best-performing neural network model can correctly classify 57.3 per cent ± 1.1 per cent. ML methods in astronomy have been receiving much interest as Vera C. Rubin Observatory prepares for first light. The observables in this study are selected to be directly comparable to early Rubin Observatory data and the prospects for running ML algorithms on the upcoming data set yields promising results.

Extremely Randomized Trees With Privacy Preservation for Distributed Structured Health Data

Artificial intelligence and machine learning have recently attracted considerable attention in the healthcare domain. The data used by machine learning algorithms in healthcare applications is often distributed over multiple sources, for instance, hospitals or patients’ personal devices. One main difficulty lies in analyzing such data without compromising patients’ privacy and personal data, which is a primary concern in healthcare applications. Therefore, in these applications, we are interested in running machine learning algorithms over distributed data without disclosing sensitive information about the data subjects. In this paper, we propose a distributed extremely randomized trees algorithm for learning from distributed data with privacy preservation. We present the implementation of our technique (which we refer to as <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${k}$ </tex-math></inline-formula> -PPD-ERT) on a cloud platform and demonstrate its performance based on medical data, including Heart Disease, Breast Cancer, and mental health datasets (Depresjon and Psykose datasets) associated with the Norwegian INTROducing Mental health through Adaptive Technology (INTROMAT) project.

Machine Learning at the Edge: A Data-Driven Architecture With Applications to 5G Cellular Networks

The fifth generation of cellular networks (5G) will rely on edge cloud deployments to satisfy the ultra-low latency demand of future applications. In this paper, we argue that such deployments can also be used to enable advanced data-driven and Machine Learning (ML) applications in mobile networks. We propose an edge-controller-based architecture for cellular networks and evaluate its performance with real data from hundreds of base stations of a major U.S. operator. In this regard, we will provide insights on how to dynamically cluster and associate base stations and controllers, according to the global mobility patterns of the users. Then, we will describe how the controllers can be used to run ML algorithms to predict the number of users in each base station, and a use case in which these predictions are exploited by a higher-layer application to route vehicular traffic according to network Key Performance Indicators (KPIs). We show that the prediction accuracy improves when based on machine learning algorithms that rely on the controllers’ view and, consequently, on the spatial correlation introduced by the user mobility, with respect to when the prediction is based only on the local data of each single base station.

An NB-IoT-Based Edge-of-Things Framework for Energy-Efficient Image Transfer.

Machine Learning (ML) techniques can play a pivotal role in energy efficient IoT networks by reducing the unnecessary data from transmission. With such an aim, this work combines a low-power, yet computationally capable processing unit, with an NB-IoT radio into a smart gateway that can run ML algorithms to smart transmit visual data over the NB-IoT network. The proposed smart gateway utilizes supervised and unsupervised ML algorithms to optimize the visual data in terms of their size and quality before being transmitted over the air. This relaxes the channel occupancy from an individual NB-IoT radio, reduces its energy consumption and also minimizes the transmission time of data. Our on-field results indicate up to 93% reductions in the number of NB-IoT radio transmissions, up to 90.5% reductions in the NB-IoT radio energy consumption and up to 90% reductions in the data transmission time.

Analysis of Machine Learning Algorithms for Anomaly Detection on Edge Devices.

The Internet of Things (IoT) consists of small devices or a network of sensors, which permanently generate huge amounts of data. Usually, they have limited resources, either computing power or memory, which means that raw data are transferred to central systems or the cloud for analysis. Lately, the idea of moving intelligence to the IoT is becoming feasible, with machine learning (ML) moved to edge devices. The aim of this study is to provide an experimental analysis of processing a large imbalanced dataset (DS2OS), split into a training dataset (80%) and a test dataset (20%). The training dataset was reduced by randomly selecting a smaller number of samples to create new datasets Di (i = 1, 2, 5, 10, 15, 20, 40, 60, 80%). Afterwards, they were used with several machine learning algorithms to identify the size at which the performance metrics show saturation and classification results stop improving with an F1 score equal to 0.95 or higher, which happened at 20% of the training dataset. Further on, two solutions for the reduction of the number of samples to provide a balanced dataset are given. In the first, datasets DRi consist of all anomalous samples in seven classes and a reduced majority class (‘NL’) with i = 0.1, 0.2, 0.5, 1, 2, 5, 10, 15, 20 percent of randomly selected samples. In the second, datasets DCi are generated from the representative samples determined with clustering from the training dataset. All three dataset reduction methods showed comparable performance results. Further evaluation of training times and memory usage on Raspberry Pi 4 shows a possibility to run ML algorithms with limited sized datasets on edge devices.

Boosted Genetic Algorithm Using Machine Learning for Traffic Control Optimization

Traffic control optimization is a challenging task for various traffic centers around the world and the majority of existing approaches focus only on developing adaptive methods for normal (recurrent) traffic conditions. Optimizing the control plans when severe incidents occur still remains an open problem, especially when a high number of lanes or entire intersections are affected. This paper presents a novel methodology for optimizing the traffic signal timings in signalized urban intersections, under non-recurrent traffic incidents. With the purpose of producing fast and reliable decisions, we combine the fast running Machine Learning (ML) algorithms and the reliable Genetic Algorithms (GA) into a single optimization framework. Firstly, we deploy a typical GA algorithm by considering the phase duration as the decision variable and the objective function as the total travel time in the network. We fine tune the GA for crossover, mutation, fitness calculation and obtain the optimal parameters. Secondly, we train several regression models to predict the total travel time in the studied traffic network, and select the best performing model which we further hyper-tune. Lastly, we propose a new algorithm BGA-ML combining the GA algorithm and the extreme-gradient decision-tree (XGBT), which is the best performing regression model, together in a single optimization framework. Comparison and results are generated by two experiments (one synthetic and one from real urban traffic network) and show that the new BGA-ML is much faster than the original GA algorithm and can reduce the total travel time by almost half when used under incident conditions.

Otel Oda Fiyatlarını Açıklamada Makine Öğrenmesi Algoritmalarının Kıyaslanması (Comparison of Machine Learning Algorithms in Explaining of Hotel Room Prices)

Purpose: The aim of the study is to analyze some of the main machine learning algorithms using big data obtained by web mining and to test the performance of these algorithms to explain hotel room prices. Thus, it is the determination of the model that best explains the hotel room prices. Design/methodology/approach – Web mining/scraping method was used to obtain research data. The target website was scanned for about six months with the help of an algorithm, and the data obtained from 6558 accommodation facilities were used in analysis. The second part of the research consists of statistical analysis and comparison of machine learning algorithms. Python programming language was used for analysis and implementation of algorithms. Pandas, numpy libraries for data processing; seaborn, matplotlib for graphics and visualization; scikit-learn is used to run machine learning algorithms. After the analysis, a model was created by logistic regression, which was thought to be the most suitable for the data. Results: It is seen that the compared Random Forest and Decision Tree algorithms both explain the data set at a rate of approximately 99.89%, so the tree/branching has been successful. The KNN algorithm achieved the highest performance with a classification of three clusters at 62.12%. Logistic Regression, Stochastic Gradient Decent and Support Vector Machines using the linear classification method obtained the highest score with 39.13% logistic regression method. In the model created by logistic regression, the score given to the hotel by the guests, the rank of the hotel among other hotels in the region, the type of the hotel and the city in which it is located were found to be statistically significant (p <0.05). Discussion: As a result of the research, machine learning algorithms were compared using the hotel room price data that obtained from web mining/scraping. It was also found important information about the hotel room rates in Turkey. A regression model has revealed which of the 44 independent variables are significant in explaining the hotel room price.