Traffic flow on long-distance highways, especially at sections with multi-interchanges and ramps, exhibits nonlinear trends affected by long-term and short-term spatiotemporal dependencies, resulting limited fitting capabilities for the major applied spatiotemporal forecasting models in use. This paper tackles this challenge by integrating an incremental gate state output decomposition (IGOD) mechanism into the recurrent neural network (RNN) model framework, accounting for the interdependencies of spatiotemporal traffic data. The proposed method improves the ability of the RNN model to estimate traffic data series by segmenting consecutive time intervals and accumulating incremental changes across these time intervals, allowing for more precise traffic predictions. This study also explores how threshold amplitudes affect prediction effectiveness. We applied it to real traffic data from segment k602+630 to k625+420 on the Changjiu Highway. The results demonstrate that the proposed model consistently exhibits robustness, with variations in threshold magnitude having little impact on its prediction accuracy.

The pupil, iris, vitreous, and retina are parts of the eye, where any defect due to physical damage or chronic diseases to these parts of the eye can lead to partial vision loss or complete blindness. Changes in retinal structure due to diabetes or high blood pressure lead to diabetic retinopathy (DR). The early diagnosis of DR using computer-aided automated tools is possible due to tremendous advancements in machine and deep learning models in the last decade. Devising and implementing innovative deep learning models for retinal structural analysis is crucial to the early diagnosis of DR and other eye diseases. In this work, we have developed a new approach, which involves the development of a lightweight convolutional neural network (CNN)-based model for segmentation of retinal vessels and a mobile application for DR grading. This paper covers the development process of an Android application that leverages the power of CNN-based deep learning model to detect DR regardless of its stage. To achieve this, two models have been created and compared, the best one having an accuracy of 96.72 %. An Android application has then been developed, that makes calls to this model and then displays the results on screen with a simple-to-understand interface developed using the Kivy framework.

This study is an exploration of the design of the state-of-the-art intent-based networking (IBN) model. In IBN, communication means are initialised by user’s (herein IT staff, not end-user) input of requirements and not instructions. Thus, allowing the self-organisational abilities of the network to set communication paths. Through research of academic studies and standardisation drafts we conduct IBN structure. We determined the need for change in the design. The current IBN model detains its adaptation as network assurance requirements of ensuring network security and scalability, and continuity are unfulfillable via conduct of network analysis and track of intent drift. We propose two submodels - one for autonomous networks and one for supervised networks.

In recent years, interest in memelements, including memcapacitors, has increased significantly following the realisation of memristors. This paper presents the design and implementation of a memcapacitor circuit based on operational transconductance amplifiers (OTAs). The proposed design is structured as a mutator circuit, where the second stage functions as a memristor, ultimately transforming the circuit into a memcapacitor emulator. The emulator features electronic tunability, which allows the charge value of the memcapacitor to be adjusted by modifying the capacitor in the mutator stage. The charge value of the memcapacitor can also be adjusted by varying the transconductance gm value of the OTA active element. Additionally, the operational frequency of the memcapacitor can be varied by altering the capacitor in the second stage. An adaptive learning circuit based on the memcapacitor emulator is demonstrated to validate the circuit performance. The time response obtained when a sine signal is applied to the memcapacitor circuit, the input voltage-charge relationship, and the charge-time response obtained when a square wave is used to demonstrate its memory characteristics are provided. All simulations were conducted using LTSpice with Taiwan Semiconductor Manufacturing Company (TSMC) 0.18 μm complementary metal oxide semiconductor (CMOS) process parameters. The results corroborate the effectiveness of the circuit, highlighting its potential for advanced electronic applications.

The implementation of biological neuron models plays an important role in understanding the functionality of the brain. Generally, analog and digital methods are preferred during implementation processes. The Raspberry Pi (RPi) microcontroller has the potential to be a new platform that can easily solve complex mathematical operations and does not have memory limitations, which will take advantage while realizing biological neuron models. In this paper, Hodgkin-Huxley (HH), FitzHugh-Nagumo (FHN), Morris-Lecar (ML), Hindmarsh-Rose (HR), and Izhikevich (IZ) neuron models have been implemented on a standard-equipped RPi. For the numerical solution of each neuron model, the one-step method (4th order Runge-Kutta (RK4), the new version of Runge-Kutta (RKN)), the multi-step method (Adams-Bashforth (AB), Adams-Moulton (AM)), and predictor-corrector method (Adams-Bashforth-Moulton (ABM)) are preferred to compare results. The implementation of HH, ML, FHN, HR, and IZ neuron models on RPi and the comparison of numerical models RK4, RKN, AB, AM, and ABM in the implementation of neuron models were made for the first time in this study. Firstly, MATLAB simulations of the various behaviors belonging to the HH, ML, FHN, HR, and IZ neuron models were completed. Then those models were realized on RPi and the outputs of the models are experimentally produced. The errors are also presented in the tables. The results show that RPi can be considered as a new alternative tool for making complex neuron models.

Chronic venous insufficiency (CVI) is a serious disease characterised by the inability of the veins to effectively return blood from the legs back to the heart. This condition represents a significant public health issue due to its prevalence and impact on quality of life. In this work, we propose a tool to help doctors effectively diagnose CVI. Our research is based on extracting Visual Geometry Group network 16 (VGG-16) features and integrating a new classifier, which exploits mean absolute deviation (MAD) statistics to classify samples. Although simple in its core, it outperforms state-of-the-art method which is known as the CVI-classifier in the literature, and additionally it performs better than the methods such as multi-layer perceptron (MLP), Naive Bayes (NB), and gradient boosting machines (GBM) in the context of VGG-based classification of CVI. We had 0.931 accuracy, 0.888 Kappa score, and 0.916 F1-score on a publicly available CVI dataset which outperforms the state-of-the-art CVI-classifier having 0.909, 0.873, and 0.900 for accuracy, Kappa score, and F1-score, respectively. Additionally, we have shown that our classifier has a generalisation capacity comparable to support vector machines (SVM), by conducting experiments on eight different datasets. In these experiments, it was observed that our classifier took the lead on metrics such as F1-score, Kappa score, and receiver operating characteristic area under the curve (ROC AUC).

This paper presents a comprehensive study on the application of deep learning techniques to accurately detect sleep apnea. The study leverages a dataset obtained from the Sleep Laboratory of the Department of Chest Diseases of Yozgat Bozok University, with the aim of developing an effective decision support system capable of identifying cases of sleep disorders with high accuracy. The proposed methodology focusses on the use of deep neural networks (DNNs) to enhance the accuracy and reliability of sleep apnea detection. By employing meticulous data collection, preprocessing, and analysis, the study demonstrates the potential of DNNs to capture intricate and high-dimensional features within complex sleep data, allowing precise and reliable diagnosis. The experimental results showcase the effectiveness of the proposed DNN-based classifier design, achieving an accuracy of 96.48 %. The study’s contributions lie in the enhancement of sleep disorder diagnosis through the integration of deep learning techniques, offering promising implications for clinical practice. Early detection of sleep disorders has the potential to significantly improve patient outcomes and overall quality of life and lays the foundation for further advancements in the field of sleep medicine.

Induction machines (IMs) are crucial to driving auxiliary machinery and devices of ships, such as pumps, fans, winches, and elevators, which are essential for maintaining the operational functions of a ship and are characterised by various types of loads. This is of critical importance because high surges of starting current can cause instabilities and fluctuations in a ship’s power system, directly affecting the safety and efficiency of ship operations. This paper provides a comprehensive review of analytical expressions for modelling the start-up time of directly started IMs under different ship operational conditions (no-load, linear load, fan load, and gravitational load). Validation of the analytical models was performed by comparing the speed-time characteristics obtained from the experimental measurements with the corresponding ones obtained by simulations in a MATLAB Simulink environment. The observed 1.5 kW IM reached a steady state for 0.1356 seconds when driving the load with a fan characteristic. However, when subjected to linear and gravitational loads, the IM requires longer times to reach a steady state - 0.1400 and 0.1606 seconds, respectively. The results of the simulations and experimental tests highly corresponded with the analytical predictions, confirming the reliability and practical applicability of the analytical models.

A non-fungible token (NFT) is a kind of digital asset that signifies ownership or proof of authenticity of a special good or piece of material, such as artwork, music, films, or tweets. This study investigates how a deep convolutional generative adversarial network (DCGAN) can be used to create distinctive pictures of Cryptopunks that can be converted into NFTs. Cryptopunks, a pioneering form of NFTs, were introduced on the Ethereum blockchain in 2017 as part of a social experiment. In the NFT community, they have since grown in popularity as collectibles. To create brand-new, previously undiscovered characters, we trained a model on a dataset of existing Cryptopunks using the DCGAN architecture. In an effort to raise the calibre of the images produced, we tested various hyper settings and layer combinations. We also assessed the created images using a variety of criteria, such as the inception score and Fréchet inception distance, to make sure they were distinctive and of high calibre. Our experiments yielded a 15 % increase in the inception score and a 20 % decrease in the Fréchet inception distance, showing that our DCGAN model produces images that are more visually appealing and closer in quality to real Cryptopunks. These results highlight the effectiveness of our machine learning algorithms in improving the quality and uniqueness of NFT assets.

The global demand for energy has increased exponentially over the years. To reduce the dominance of fossil fuels in energy production, there has been a shift towards energy production models based on renewable sources. In the design of hybrid energy systems, it is essential to keep investment costs low while ensuring the security of the energy supply by meeting the consumer’s energy demands without interruption. The success of a good energy production model can be directly associated with the results of load estimation. The primary objective of this research is to predict the electricity demand for the Göksun district until 2028, utilising a data set that encompasses electricity usage from 2019 through the first four months of 2024 for the Göksun district in Kahramanmaraş. This endeavour includes the application of various machine learning (ML) paradigms (long short-term memory (LSTM), gated recurrent unit (GRU), convolutional neural network (CNN)-LSTM, support vector regression (SVR)) to produce load forecasting outcomes and to engineer an optimally performing hybrid system. On evaluation of the performance metrics derived from the experimental data, it has been established that the LSTM model outperforms other methodologies, yielding more favourable results. The simulation studies of the designed hybrid system were conducted using the hybrid optimisation model for electric renewables software (HOMER Pro), demonstrating improvements in both economic and environmental parameters. Our study is unique in that it is the first to utilise a data set specific to the Göksun region and to model predictions obtained from this data set using HOMER software.