Journal of Artificial Intelligence, Applications and Innovations

In contemporary times, a substantial number of financial transactions and monetary transfers take place on the Internet and within electronic environments, thereby incentivizing fraudsters to infiltrate this domain. Consequently, the identification of individuals' identities in electronic service provision is exceedingly vital and crucial. This article aims to fraud detection in the banking system and present an optimal method utilizing artificial intelligence tools and model evaluation on the bank information of the Development and Cooperation Cooperative. In the initial phase, a gradient boosting algorithm, chosen for its high computational speed, is employed to train on a set of input data to identify and classify patterns of suspicious behaviors. In the second phase, an algorithm based on gradient boosting is utilized to refine results and optimize accuracy. To evaluate this approach, real data from a bank is employed, and the obtained results demonstrate that this method significantly enhances the speed and accuracy of fraud detection.

Federated learning is a novel way of training machine learning models on data that is distributed across multiple devices, such as smartphones and IoT sensors, without compromising privacy, efficiency, or security. However, federated learning faces a significant challenge when the data on each device is not independent and identically distributed (non-IID), which means that the data may have different distributions, sizes, or qualities. non-IID data is a major challenge for federated learning, as it affects the accuracy and participation of the local devices. Most existing methods focus on improving the model, algorithm, or framework of federated learning to deal with non-IID data. However, there is a lack of systematic and up-to-date reviews on this topic. In this paper, we survey different approaches to address the challenge of non-IID data in Vertical Federated Learning (VFL) and Horizontal Federated Learning (HFL). We organize the existing literature based on the perspective of the researcher and the sub-tasks involved in each approach. Our goal is to provide a comprehensive and systematic overview of the problem and its solutions.

With the expansion of social networks, their utilization in digital advertising has become a key factor in shaping public opinion and driving advertising campaigns —coordinated and targeted series of interactions between users on a specific topic. Properly directing these campaigns can focus many individuals on a particular subject, thereby creating effective campaigns. In this research, we introduce a method for developing campaigns suitable for digital advertising on Twitter (X). Users can leverage hashtags, tweets, comments, retweets, and other features of Twitter for a specified topic to build a campaign. This method engages known bot-connected user accounts on Twitter to interact with one another on a topic, generating initial attention and kickstarting the campaign. By then identifying influential users in that area and interacting with them, the campaign is further developed over time. To evaluate the performance of the proposed method, we considered two factors: the number of users involved in the campaign and the relevance of the selected content to the topic. We conducted this experiment with 50 bot-connected user accounts on Twitter. The results revealed that, through 116,594 interactions and receiving 246 responses from non-bot users, the proposed method was able to engage the audience within 5 days.  These results demonstrate that our approach succeeded in attracting users and receiving feedback by publishing relevant content, suggesting its potential for real-world success.

The development of Artificial Intelligence (AI) models with an accurate prediction of human facial expression has become a significant challenge for the cases in which masks and sunglasses cover critical facial areas. Given that a substantial portion of human interactions involves non-verbal communication, accurately detecting human emotions such as anger, fear, disgust, happiness, sadness, and surprise would benefit a wide range of applications, from security assessments to psychological treatments. As a workaround, the current study explores the performance of a novel multi-channel arrangement comprised of a Haar-wavelet, Histogram of Oriented Gradients (HOG), and grayscale filters to improve the predictions of deep Convolutional Neural Network (CNN) on occluded results. This study uses the FER-2013 dataset and produces occluded samples by applying a virtual mask that covers almost 55% of facial areas comprising the mouth, lips, and jaw locations. Further investigations, including the impact of each filter, utilizing pre-trained models on occluded samples (transfer learning), and comparison to prior models are also carried out. The proposed approach yields an accuracy rate of 71% for non-occluded and 66% for the occluded samples, which are 6% to 11% higher than the base model. Further transfer learning technique increases the accuracy metrics by 18%, indicating that non-occluded pre-trained models can reveal a broader range of features and their relation, which to some extent compensates for the removed features due to the occlusion. These results suggest the potential capabilities of the proposed technique for similar imaging applications.

Quantum computing, with its promise of exponential speedups and novel algorithms, has emerged as a revolutionary technology in various computational fields, including image processing. This survey provides a comprehensive review of the applications of quantum computing in image processing, exploring how quantum algorithms and hardware can address classical challenges such as image compression, enhancement, pattern recognition, and image recovery. The paper delves into the theoretical foundations of quantum computing, discusses current advancements, and compares quantum methods with traditional approaches. Moreover, it identifies the key challenges, such as scalability and hardware limitations, that currently hinder the widespread adoption of quantum techniques in image processing. Finally, the survey outlines future research directions, highlighting the potential for further integration of quantum computing with advanced image processing technologies like deep learning. This work aims to serve as a foundational reference for researchers and practitioners interested in the intersection of quantum computing and image processing.

The rise of large language models (LLMs) has transformed numerous natural language processing (NLP) tasks, yet their performance in low and mid-resource languages, such as Farsi, still lags behind resource-rich languages like English. To address this gap, we introduce Naab, the largest publicly available, cleaned, and ready-to-use Farsi textual corpus. Naab consists of 130GB of data, comprising over 250 million paragraphs and 15 billion words. Named after the Farsi word ناب (meaning "pure" or "high-grade"), this corpus is openly accessible via Hugging Face, offering researchers a valuable resource for Farsi NLP tasks. In addition to naab, we provide naab-raw, an unprocessed version of the dataset, along with a pre-processing toolkit that allows users to clean their custom corpora. These resources empower NLP researchers and practitioners, particularly those focusing on low-resource languages, to improve the performance of LLMs in their respective domains and bridge the gap between resource-rich and resource-poor languages.

This study focuses on predicting cardiovascular disease (CVD) mortality using various machine learning (ML) techniques. A diverse set of parameters from different categories within the Sleep Heart Health Study (SHHS) dataset is leveraged, and ML techniques including LR, KNN, SVM, RF, ETC, and SGD, are employed. To ensure the reliability of these techniques, 10-fold cross-validation is applied. Furthermore, the mutual information technique with K-fold stratified cross-validation is used to determine feature importance, enhancing the model’s interpretability. The proposed approach predicts CVD mortality over a 10 to 15-year period and aims to identify influential parameters to facilitate timely interventions and lifestyle improvements for patients, ultimately contributing to an increased lifespan. Among the algorithms, KNN outperforms others, achieving an accuracy of 77%, an F1-score of 77%, an AUC of 79%, a sensitivity of 77.34%, and a specificity of 76.56%.

Reasoning refers to the extraction of information from a graph in an embedded space through link prediction. Link prediction involves identifying the missing edge between a pair of entities. While graph features are utilized to locate these missing edges, similarity-based methods often fail to consider all of the graph features. Multi-view methods leverage multiple perspectives, each offering a different aspect of data analysis. In this paper, we present a novel approach called the Multi-View Probabilistic Latent Semantic Analysis (PLSA) based PLSA algorithm. This algorithm calculates three distinct views, enabling a comprehensive analysis of the underlying data. The first one refers to the viewing probability of a node in the head, tail, or relation separately. The second view highlights the significance of a suggested tail in information representation, while the third view evaluates the quality of information flow within a set comprising the head, relation, and tail. These three views are combined to derive the optimal score function by PLSA method. Test results indicate that the proposed method ensures that the correct solution lies within an acceptable range, with a hit rate exceeding 40% in the Freebase dataset. Experimental results further demonstrate the effectiveness of the proposed algorithms compared to state-of-the-art methods.