سید عنایت الله علوی

Some companies manipulate financial statements to users and commit fraud. Therefore, effort to detect fraud is essential. Meanwhile, data mining techniques have increasingly popular. This article aimed to use an advanced model to detect fraudulent financial statements and compare it with the other methods. Crowd optimization algorithms have been considered to solve many optimization problems, but so far, they have not been used in fraud detection research to determine the optimal value of SVM parameters and optimize ANN architecture. In this research, for the first time, the PSO algorithm was used as one of the best innovative optimization algorithms for these optimizations due to its memory and high convergence speed, as well as having solutions for exiting from local optimal points and cooperation and information sharing between particles to detect fraud. For this purpose, the financial statements of companies admitted to the stock exchange from 2017 to 2023 were reviewed. The findings showed that the SVM-PSO method, with 89.86%accuracy, compared to the ANN-PSO method, with 80.43%accuracy, and the LR method, with 69.57%accuracy, performs better in identifying suspected fraudulent financial statements. Combining the PSO algorithm with the SVM method has proven superior to other methods due to SVM's high ability to reduce false negatives and PSO's ability to fine-tune its parameters. This combination can be used for high-accuracy financial statement fraud detection.

Automating the processing of videos in applications such as surveillance, sport commentary and activity detection, human-machine interaction, and health/disability care is crucial to their correct functioning. In such video processing tasks, recognition of various human actions is a pivotal component for the correct understanding of videos and making decisions upon it. Accurately recognizing human actions is a complex process, demanding high computing capabilities and intelligent algorithms. Several factors, such as object occlusion, camera movement, and background clutter, further challenge the task and its accuracy, essentially leaving deep learning approaches the only viable option for properly detecting human actions in videos. In this study, we propose CoReHAR, a novel Human Action Recognition method that employs both deep Convolutional and Recurrent neural networks on raw video frames. Using the pre-trained ResNet152 CNN, deep features are initially extracted from video frames. The sequential information of the frames is then learned using DB-LSTM RNN. Multiple stacked layers in forward and backward passes of the DB-LSTM provide increased network depth for higher accuracy. A number of techniques are also applied to improve CoReHAR’s processing speed on heterogeneous GPU-enabled systems. The proposed method is evaluated using PyTorch, and is compared to the state-of-the-art methods, showing a considerable efficiency increase, with nearly 95% recognition accuracy measured as an average over all splits of the challenging UCF101 dataset.

In this paper, a fuzzy ontology for Semantic Sensor Networks (SSN) is proposed for smart homes in two phases. In the first phase, using the WordNet ontology, the location and type of an object is identified with the aid of a graphical interface. This object and its synonyms are added to the list of the known objects set. Succeeding, the relation of the object with other groups is assessed based on a similarity measure in addition to using the fuzzy ontology. In the second phase, sensors with erroneous information are identified and pruned by finding a relationship between some specific factors. To this end, temperature, moisture and light are considered and the Adaptive Neuro-Fuzzy Inference System (ANFIS) is incorporated. The proposed method is implemented using some parts of the Wikipedia database and the WordNet dictionary. The first phase of the proposed method is tested with several sample requests and the system shows favorable results on finding the original group (and other related groups) of the request. For training the neural network in the second phase, the Intel lab Dataset is used. Results of this phase show that the neural network can predict the temperature and moisture factors with low error, while the light factor has more error in prediction