نشریه سیستم های پردازشی و ارتباطی چند رسانه ای هوشمند
سال پنجم شماره 2 (پیاپی 16، تابستان 1403)

In the domain of financial forecasting, machine learning (ML) models have garnered significant attention in recent years. One prominent application lies in the cryptocurrency market, where intelligent trading bots facilitate a substantial portion of daily transactions.

This paper investigates the efficacy of ML-based methods for identifying optimal trade entry points. Specifically, we employ the Relative Strength Index (RSI) and Simple Moving Average (SMA) indicators to extract a set of trend and crossover features. Subsequently, these features are utilized to train multilayer neural network, support vector machine, and nearest neighbor learning models. The performance of each model is then evaluated using digital currency market data encompassing the first seven months of 2023 for a variety of cryptocurrencies.

Our findings demonstrate that, firstly, the extracted features exhibit promising efficacy. Secondly, the nearest neighbor model achieved the highest profitability during the evaluation period compared to the other investigated models.

In the research conducted in this field, technical indicators are often used directly in market forecasting but in the proposed method of this article, instead of directly using the values of the indicators as the input of the classification methods, trend behaviors and their intersections have been used. In the continuation of this research, it is possible to determine the best exit points from the transaction with the help of machine learning and the use of volume indicators in the learning process.

Inadequate risk assessment of policyholders is a significant challenge in the health insurance industry. This research proposes a customer validation model for health insurance plans, focusing specifically on employees of the East Iran Oil Company.

In this study, a Multi-Layer Perceptron (MLP) neural network with six steps is employed for customer validation. Weight training, a crucial step in neural network implementation, determines the influence of explanatory variables on the model's output. The trained model is then used for validation.

Validation results indicate that health insurance claim variables and specific disease variables have the highest impact on unhealthy customer classification. Notably, the validation process identified approximately 1.8% of the population as "unhealthy." This seemingly small group accounts for a disproportionately high 17.47% of the company's total health insurance claims, despite currently being classified and charged premiums as healthy individuals.

The proposed validation model offers a practical approach for insurance companies to assess customer risk profiles and tailor premiums accordingly. This approach promotes a more equitable and sustainable insurance system. JEL classification: C52, I13, C45

Since the emergence of mobile apps, user reviews have been of great importance for app developers, as they contain users’ sentiment, bugs and new requests, user feedback has been valuable to app developers. Due to the large number of reviews, analysing them is a difficult, time-consuming and error-prone task. So far, many works have been done in the field of classifying reviews; But in none of them, it is not clear which category the users paid more attention to after categorization.

In this article, an algorithm called UAR(User Attention Rate) is presented based on the date of reviews and their number to measure the level of attention of users to the category of bug report or feature request. To test the proposed algorithm, the reviews of WhatsApp, NextCloud, Mozilla Firefox and VLC Media Player apps were extracted by the web crawler for 30 days, and then the reviews were given to 4 software experts after classification. Next, after analyzing the reviews by 4 experts, the level of attention of users to the reviews is calculated by them. Then, using the proposed method, amount of users' attention to reviews is measured.

It was found that the proposed algorithm able to have acceptable results in calculating the amount of users' attention to reviews compared to the values calculated by experts. So that the difference in the attention rate calculated by experts according to the proposed algorithm was 2.73%.

Due to the large number of reviews posted for apps, it is difficult and time-consuming to review them by the development team. Having a method for User Attention Rate could save the development team’s time and help implement new features in the app, fix bugs and make the app successful. So far, many methods have been presented for automated, however, most of them have focused on invalid old features or ignored new features. A new algorithm is presented to calculate the users' attention rate to the categories of feature requests and bug reports. By using the relevant algorithm, the development team will understand that the user wants to fix the bugs related to the application or add new features to the application.

The increase in the number of PEVs in recent years requires solving the challenges related to them, such as providing charging and discharging infrastructure and managing load changes caused by charging cars on the distribution network. PEVs spend a significant amount of time in the parking mode. Therefore, the use of electric energy stored in PEVs batteries using the vehicle-to-grid capability plays a significant role in solving the challenges of the distribution network in the coming years. Charging parked PEVs during off-peak periods and discharging them during peak times is an interactive strategy between electric vehicles and the distribution network in order to improve or solve challenges. As a result of these mutual effects, in the planning and management of the charging of these PEVs, the indicators of the distribution network such as energy losses, voltage profile, reliability are always considered. Therefore, it is very important to allocate a suitable PL based on technical and economic issues. Reducing the unsupplied energy in the distribution network, by using the energy stored in the batteries of PEVs, increases the reliability of the network.

In this research, a model based on mixed-integer linear programming is proposed to determine optimal location and capacity of PEV’s parking lots with the aim of maximizing the profit of PL owners Also, a method for measuring the reliability of the distribution network has been proposed. In the proposed model, the goals of trips, the number of PEVs and their arrival and departure times in different areas, based on the type of trip, are taken into account. The constraints of the distribution network are also taken into account using linearized equations for distribution power flow. The suggested model has been implemented in a distribution system with 37 buses, which includes four different areas in terms of the type of vehicle travel.

The results show the effectiveness of the proposed model for locating electric vehicle’s PLs and measuring the reliability of the network.

Three case studies were evaluated in this paper. In the first case study, the state of losses and the reliability of the network under test, along with verification, have been checked before the parking lot is set up; In the second case study, the primary goal of the presented model is to maximize the profit of the parking lot investor, and in this case, the maximum profit that can be obtained is calculated as 3.735 million dollars, and also in the second step, the reliability of the network is measured under The study found that the parking lots in the system were effective in increasing the reliability of the network by 1.4%. In the third case study, it has been determined that moving the location of the parking lots can have a positive effect on the reliability of the network.

Medical imaging modalities play a crucial role in the diagnosis, staging, treatment, and monitoring the treatment process of diseases. Photoacoustic imaging has emerged as a recent addition to medical research, offering non-invasiveness, enhanced safety due to non-ionizing radiation, and ease of application in various medical fields. This imaging technique combines ultrasound and optical technologies, providing the significant advantage of high-contrast optical imaging and high-resolution acoustic imaging. This article reviews the fundamental principles of photoacoustic imaging and its common modalities used in medical applications, including photoacoustic microscopy, photoacoustic computed tomography, and photoacoustic endoscopy, as well as the challenges in this field. Studies indicate the potential of photoacoustic technique in providing molecular, structural, and functional information from biological tissues.

In today's ever-evolving landscape of technology, the Internet of Things (IoT) has emerged as a transformative force, interconnecting countless smart devices that permeate our daily lives. From smart homes and cities to industrial automation and healthcare, IoT has brought about unprecedented convenience and efficiency. However, this rapid proliferation of IoT devices has also given rise to significant security challenges. The IoT ecosystem encompasses a diverse array of devices, ranging from wearable fitness trackers to critical infrastructure components, all of which are susceptible to cyber threats. Unauthorized access, data breaches, and malicious attacks on IoT networks pose severe risks to data privacy, infrastructure stability, and public safety. As a result, the need for robust security measures, such as IoT Intrusion Detection Systems, has become increasingly evident. The importance of these systems cannot be overstated, as they serve as the first line of defense against a myriad of IoT-related threats. By identifying and responding to potential security breaches, IoT Intrusion Detection Systems help maintain the integrity of data, ensure the functionality of IoT devices, and preserve the trustworthiness of the entire IoT network. The limitation of resources in electronic devices of the Internet of Things has caused less attention to the security. Today, Intrusion detection systems (IDSs) are one of the most important solutions to identify all types of attacks and threats and adopt appropriate solutions to deal with them. In addition, due to the openness of the environment, the placement of devices based on the Internet of Things makes this environment more vulnerable. For this reason, providing an effective and efficient intrusion detection system can be a suitable solution for this environment.

In this article, we have presented a two-layer intrusion detection system based on KNN classification to separate malicious traffic from normal mode and multilayer perceptron artificial network to detect the type of attack. The data set used is KDD-CUP 99 data set.

The experiment results show 99.743% accuracy for the data set as well as the improvement of Accuracy, Recall, Precision, F-measure, TPR and FPR parameters. In addition, the delay time of the proposed method is improved 40% compared to the MLP-MLP method and has a 139% lower delay than the fog-free state.

The present study aims to propose two-layer hierarchical IDS based on machine learning, which detects attacks by considering the limitations of IoT resources. In order to create an efficient and accurate IDS, the combination of two improved K-nearest neighbor (KNN) algorithms and multi-layer perceptron (MLP) neural network applied in the fog and cloud to separate the attacks from normal traffic, respectively. we evaluated our proposed method using KDD-CUP 99 dataset. The results prove the improvement in accuracy, compared to the previous methods.