Transactions on Machine Intelligence2821-16935220220601A Bidirectional Long Short-Term Neural Network Model to Predict Air Pollutant Concentrations: A Case Study of Tehran, Iran637615999310.47176/TMI.2022.63ENH.Ghayoumi ZadehAssistant Prof, Department of Electrical Engineering, Faculty of Engineering, Vali-E-Asr University of Rafsanjan, Rafsanjan, Iran.A.FayaziAssistant Prof, Department of Biomedical Engineering, Meybod University, Meybod, Iran.O.Rahmani SeryasatAssistant Prof, Department of Electrical Engineering, Shams Higher Education Institute, Gorgan, Iran.0000-0002-9289-6128H.RabieeDepartment of Electrical Engineering, Karaj Branch, Islamic Azad University, Karaj , Iran.Journal Article20220113Air pollution remains one of the most pressing environmental challenges in modern urban societies, driven by rapid industrialization, accelerated urban expansion, increasing vehicular traffic, and intensified anthropogenic activities. The presence of harmful substances in the atmosphere, including sulfur dioxide (SO₂), nitrogen dioxide (NO₂), ozone (O₃), particulate matter (PM₂.₅ and PM₁₀), and carbon monoxide (CO), poses significant health risks and environmental hazards. Monitoring and forecasting pollutant concentrations are critical for effective air quality management and policy-making. This study presents a novel hybrid model based on deep recurrent neural networks, particularly the Bidirectional Long Short-Term Memory (BiLSTM) architecture, for short-term air quality prediction. The model focuses on forecasting the Air Quality Index (AQI), a composite measure influenced by several pollutants with highly nonlinear and complex behavior. Daily average concentration data for O₃, PM₂.₅, PM₁₀, NO₂, SO₂, and CO were collected from the Tarbiat Modares air quality monitoring station in Tehran’s 6th district (latitude 35.71751, longitude 51.385909) over a 15-month period between March 2018 and June 2019. The proposed BiLSTM model demonstrated superior performance compared to traditional shallow learning techniques and Multi-Layer Perceptron (MLP) networks. The regression coefficients (R²) for O₃, PM₂.₅, PM₁₀, NO₂, SO₂, and CO were 0.87, 0.62, 0.84, 0.67, 0.75, and 0.72, respectively. These results highlight the robustness and reliability of the BiLSTM-based approach for accurately predicting pollutant concentrations, thereby supporting timely decision-making in environmental health management.https://www.tmachineintelligence.ir/article_159993_f4151395b07c9e93d1e4593c17bfe554.pdfTransactions on Machine Intelligence2821-16935220220601Balancing Unicycle Travelling on an Inclined Surface778615999410.47176/TMI.2022.77ENE.Kouhi RonaghiElectrical & Electronic Engineering Department, Shahed University, Tehran, IranS.SeyedtabaiiElectrical & Electronic Engineering Department, Shahed University, Tehran, IranJournal Article20220119This study focuses on the control of a self-balancing unicycle robot equipped with a roll stabilization mechanism and a single drive wheel responsible for both maintaining upright posture and tracking a predefined path. To achieve stable and responsive behavior, Linear Quadratic Regulators (LQRs) are designed and implemented for roll, pitch, and trajectory tracking control loops. While conventional LQR controllers are generally effective in attenuating external disturbances such as road inclines, they may not fully compensate for the dynamic variations induced by significant slope changes. To address this limitation, a gain-scheduled LQR strategy is proposed, in which controller gains are adaptively adjusted based on varying road inclinations. The road slope alters the dynamic behavior of the unicycle system and introduces external forces that can degrade performance if not adequately compensated. The proposed gain scheduling approach enhances the system's adaptability and robustness, ensuring more accurate path tracking and upright stability under non-uniform terrain conditions. A comparative analysis is conducted between the gain-scheduled LQR and a standard fixed-gain LQR design. Simulation results demonstrate the effectiveness of the proposed method in improving the robot’s performance across varying slope conditions, validating its potential for real-world application in unstructured or sloped environments.https://www.tmachineintelligence.ir/article_159994_1cb5308f38d6887ad9afbf9fa205b0d5.pdfTransactions on Machine Intelligence2821-16935220220601Landmine Detection by Correlation Method in Different Environments879615999510.47176/TMI.2022.87ENA.GharamohammadiDepartment of Electrical Engineering, Amirkabir University of Technology, Tehran, IranY.NorouziDepartment of Electrical Engineering, Assistance prof. of Electrical Eng, Amirkabir University of Technology, Tehran, IranJournal Article20220203This study aims to detect landmines in various environments by analyzing the scattering parameters obtained via Ground Penetrating Radar (GPR). The GPR system captures the scattering parameter, which serves as the key indicator of subsurface anomalies. A reference signal is first acquired from a simplified environment where a landmine is embedded in isolation. Signals from more complex or cluttered environments are then measured and compared against this reference. The presence of a landmine alters the scattering characteristics of the medium, influencing the measured parameter. By evaluating the similarity between the reference signal and the signals collected from the target environment, it is possible to infer the existence of a landmine. This similarity is quantified using a correlation function, which effectively highlights matching patterns. The strength of this approach lies in the distinctive and invariant nature of the scattering parameter, which provides a reliable basis for detection. The proposed method offers a practical and efficient solution for landmine detection, especially in scenarios where signal clarity and robustness are essential. Its reliance on scattering parameter uniqueness ensures consistent performance, making it a promising tool for real-world applications in humanitarian demining and subsurface anomaly detection.https://www.tmachineintelligence.ir/article_159995_1709ffe1eb141260b2bc38d76aec5178.pdfTransactions on Machine Intelligence2821-16935220220601Improved Recommender Systems Using Data Mining9711416234110.47176/TMI.2022.97ENH.AviniDepartment of Computer Engineering, Faculty of Technology and Engineering, Yasouj branch, Islamic Azad University, Yasouj, Iran0009-0000-4872-3185Z.Mirzaei ZavardJaniDepartment of Computer Engineering, Faculty of Technology and Engineering, Qeshm Branch International University, Qeshm, IranA.AvinIDepartment of Accounting, Faculty of Accounting, Yasouj branch, Islamic Azad University, Yasouj, Iran.Journal Article20220223Today, in order to buy goods through the Internet, every company or production organization has an internal commercial software site based on which it offers its products and services to customers. To ensure that the user has the ability to provide a suitable proposal for a request or to solve a need in the midst of a huge amount of data, recommender systems are the right solution. Different methods of providing suggestions in recommender systems are divided into eight methods according to the data mining of the classification of methods. In each method, in these systems, the necessary suggestions and predictions are provided to users in a special way, and the most important method is the recommender system among other methods. , is the filtering method. In this method, the number of clusters in the data set related to recommender systems is dynamically determined by c3m clustering algorithm on a data set called Movielens ml-100k, which has a data oscillator in four inputs, as well as k-means algorithm and performance optimization. It was estimated. The final clustering is done well with the help of this method, if the target user enters after the clustering operation and by matching the profile information which includes a series of items rated by similar users in the same cluster, the similar cluster search for Based on the correlation filter, which is one of the methods used by the KNN algorithm, it finds its similar cluster with each cluster head (cluster representative) and based on the items ranked in demographic information, the nearest neighbors (neighbor and similar) finds users-items and the item that has the highest rank among other users. The obtained similarity is stored in the user's top-n list and presented in the form of an offer.https://www.tmachineintelligence.ir/article_162341_19421b53fd8e412d15949716f5cea94b.pdfTransactions on Machine Intelligence2821-16935220220601A Niching Ring Topology Genetic Algorithm for Multimodal Optimization11512118150310.47176/TMI.2022.115ENH.A.BazoobandiDepartment of Computer Engineering, Esfarayen University of Technology, Esfarayen, North Khorasan, IranJournal Article20220317Multimodal optimization represents a significant and ongoing challenge within the broader field of optimization, particularly due to the presence of multiple global and local optima within a complex search space. Unlike unimodal problems that focus on a single optimal solution, multimodal problems require algorithms to locate and maintain a diverse set of high-quality solutions across various regions of the landscape. This characteristic reflects many real-world scenarios, such as engineering design, robotics, and bioinformatics, where multiple viable solutions can coexist. Traditional optimization algorithms often struggle in such settings, as they tend to converge prematurely to a single optimum and lack mechanisms for diversity preservation. In this paper, we propose a novel niching-based Genetic Algorithm (GA) tailored specifically for multimodal optimization problems. The proposed algorithm dynamically forms niches based on the spatial distribution of individuals in the population, enabling the preservation and evolution of multiple optima simultaneously. To ensure that niches are maintained effectively, the genetic operators are strategically modified to minimize disruption to niche structure during crossover and mutation. Extensive experiments conducted on standard multimodal benchmark functions demonstrate that our approach consistently outperforms existing methods in both convergence speed and solution diversity. The results validate the algorithm’s robustness and its practical potential in solving complex multimodal problems.https://www.tmachineintelligence.ir/article_181503_15ab75396b7ec6b72073a2fe5278dc20.pdfTransactions on Machine Intelligence2821-16935220220601Implementation of a System for Removing Noisy Hyperlinks: A Semantic and Relatedness-Based Approach12213818627310.47176/TMI.2022.122ENK.TaghandikiDepartment of Computer Engineering, Technical and Vocational University (TVU), Tehran, Iran0000-0003-4415-264XElnazRezaei EhsanMaster's Degree, Industrial Engineering, System Management and Productivity, Iran University of Science and Technology, Tehran, IranJournal Article20220402As the amount of data on the web increases, the web structure graph, which represents the web as a graph, is also evolving. The structure of this graph has shifted from being based on content to being non-content-based. Additionally, spam data, such as noisy hyperlinks, in the web structure graph can negatively impact the speed and efficiency of information retrieval and link mining algorithms. Previous research in this field has concentrated on eliminating noisy hyperlinks through structural and string-based methods. However, these methods may mistakenly eliminate valuable links or fail to identify noisy hyperlinks in certain situations. In this paper, we begin by constructing a data collection of hyperlinks using an interactive crawler. We then examine the semantic and relatedness structure of the hyperlinks using semantic web tools such as the DBpedia ontology. The removal process of noisy hyperlinks is performed using a reasoner on the DBpedia ontology. Our experiments demonstrate the accuracy and effectiveness of semantic web technologies in eliminating noisy hyperlinks.https://www.tmachineintelligence.ir/article_186273_19594a60d7fbabc44f6fb1a302b61cef.pdf