Inteligencia Artificial

Accurate Price prediction by Double Deep Q-Network

Mohammd Reza Feizi Derakhshi, Bahram Lotfimanesh, Omid Amani — Fri, 17 May 2024 00:00:00 +0200

For more than several decades, time series data have been in the center of attention for scholars to predict the future prices of the markets, the most fundamental and challenging of which has been the prediction of the price of the stock market. It is of great importance to note that the algorithms with the fewest errors in price predictions are more applicable. There have been more methods suggested for price prediction in the stock markets: time series data analysis, mathematical and statistical analysis, signal processing, pattern recognition and machine learning. One of the demerits of the aforementioned methods is failing to recognize sudden change of prices, in this regard, experiencing more errors is the consequence of such demerit. In this regard, to have the error solved, the DDQN algorithm, consisting of deep neural networks which includes LSTM-CNN layers, has been employed. Confronting price fluctuations, the agent has the privilege of having better performance by employing the advantages of LSTM-CNN layers. In this research, the algorithm has been carried out over Iranian Gold Market, including six various types of Gold, from 2009 to 2020. The results reveal the point that the given method is more precise in comparison with other suggested methods confronting sudden changes in prices.

A Novel Deep Learning Model for Pancreas Segmentation: Pascal U-Net

Ender Kurnaz, Rahime Ceylan, Mustafa Alper Bozkurt, Hakan Cebeci, Mustafa Koplay — Fri, 17 May 2024 00:00:00 +0200

A robust and reliable automated organ segmentation from abdomen images is a crucial problem in both quantitative imaging analysis and computer aided diagnosis. Especially, automatic pancreas segmentation from abdomen CT images is most challenging task which based on in two main aspects (1) high variability in anatomy (like as shape, size, etc.) and location across different patients (2) low contrast with neighboring tissues. Due to these reasons, achievement of high accuracies in pancreas segmentation is hard image segmentation problem. In this paper, we propose a novel deep learning model which is convolutional neural network-based model called Pascal U-Net for pancreas segmentation. Performance of the proposed model is evaluated on The Cancer Imaging Archive (TCIA) Pancreas CT database and abdomen CT dataset which is taken from Selcuk University Medicine Faculty Radiology Department. During the experimental studies, k-fold cross-validation method is used. Furthermore, results of the proposed model are compared with results of traditional U-Net. If results obtained by Pascal U-Net and traditional U-net for different batch size and fold number is compared, it can be seen that experiments on both datasets validate the effectiveness of Pascal U-Net model for pancreas segmentation.

Age-Invariant Cross-Age Face Verification using Transfer Learning

Newlin Shebiah Russel, Arivazhagan Selvaraj, Dhanya Devi S, Dhivyarupini M — Fri, 17 May 2024 00:00:00 +0200

The integration of face verification technology has become indispensable in numerous safety and security software systems. Despite its promising results, the field of face verification encounters significant challenges due to age-related disparities. Human facial characteristics undergo substantial transformations over time, leading to diverse variations including changes in facial texture, morphology, facial hair, and eyeglass adoption. This study presents a pioneering methodology for cross-age face verification, utilizing advanced deep learning techniques to extract resilient and distinctive facial features that are less susceptible to age-related fluctuations. The feature extraction process combines handcrafted features like Local Binary Pattern/Histogram of Oriented Gradients with deep features from MobileNetV2 and VGG-16 networks. As the texture of the facial skin defines the age related characteristic the well-known texture feature extractors like LBP and HoG is preferred. These features are concatenated to achieve fusion, and subsequent layers fine-tune them. Experimental validation utilizing the Cross-Age Celebrity Dataset demonstrates remarkable efficacy, achieving an accuracy of 98.32%.

FRESHNets: Highly Accurate and Efficient Food Freshness Assessment Based on Deep Convolutional Neural Networks

Fri, 17 May 2024 00:00:00 +0200

Food freshness classification is a growing concern in the food industry, mainly to protect consumer health and prevent illness and poisoning from consuming spoiled food. Intending to take a significant step towards improving food safety and quality control measures in the industry, this study presents two models based on deep learning for the classification of fruit and vegetable freshness: a robust model and an efficient model. Models’ performance evaluation shows remarkable results; in terms of accuracy, the robust model and the efficient model achieved 97.6% and 94.0% respectively, while in terms of Area Under the Curve (AUC) score, both models achieved more than 99%, with the difference in inference time between each model over 844 images being 13 seconds.

The Superiority of Fine-tuning over Full-training for the Efficient Diagnosis of COPD from CXR Images

Victor Ikechukwu Agughasi — Fri, 17 May 2024 00:00:00 +0200

This research investigates the use of deep learning for diagnosing lung diseases like Chronic Obstructive Pulmonary Disease (COPD) using Chest X-rays (CXR). The study compares the impact of deep learning on improving these diagnoses by comparing the performance of models trained from the scratch with those enhanced through fine-tuning established architectures like InceptionV3, ResNet50, and VGG-19. The study revealed that fine-tuning pre-trained models offers significant benefits: faster convergence, improved stability, and increased accuracy. Data augmentation techniques were found to be particularly useful when dealing with limited or unbalanced datasets. A custom CNN model, Iyke-Net, showed promising results when fine-tuned. Interestingly, it was observed that models using grayscale images outperformed those using colour images in disease classification, suggesting that colour information might be less critical than previously thought for certain diagnostic procedures. The study emphasizes the importance of balancing model complexity with computational efficiency and diagnostic accuracy. It advocates for refining existing deep learning models for COPD diagnosis from CXR images, paving the way for further innovations in AI-enhanced medical diagnostics.

Machine Learning-based Extrapolation of Crop Cultivation Cost

Poonam Bari, Lata Ragha — Fri, 17 May 2024 00:00:00 +0200

It is important to comprehend the relation between operational expenses such as labour, seed, irrigation, insecticides, fertilizers and manure costs necessary for the cultivation of crops. A precise cost for the cultivation of crops can offer vital information for agricultural decision-making. The main goal of the study is to compare machine learning (ML) techniques to measure relationships among operational cost characteristics for predicting crop cultivation costs before the start of the growing season using the dataset made available by the Ministry of Agriculture and Farmer Welfare of the Government of India. This paper describes various ML regression techniques, compares various learning algorithms as well as determines the most efficient regression algorithms based on the data set, the number of samples and attributes. The data set used for predicting the cost with 1680 instances includes varying costs for 14 different crops for 12 years (2010-2011 to 2021-2022). Ten different ML algorithms are considered and the crop cultivation cost is predicted. The evaluation results show that Random Forest (RF), Decision Tree (DT), Extended gradient boosting (XR) and K-Neighbours (KN) regression provide better performance in terms of coefficient of determination (R2), root mean square error (RMSE) and mean absolute error (MAE) rate while training and testing time. This study also compares different ML techniques and showed significant differences using the statistical analysis of variance (ANOVA) test. The optimal hyperparameters for the ML models are found using the gridsearchCV and randomizedsearchCV functions, which improves the model's capacity for generalisation.

Binary Classification of Skin Cancer Images Using Pre-trained Networks with I-GWO

Hadeer Hussein, Ahmed Magdy, Rehab F. Abdel-Kader , Khaled Abd El Salam — Sun, 26 May 2024 00:00:00 +0200

One of the most prevalent forms of cancer worldwide is skin cancer. Determining disease characteristics necessitates a clinical evaluation of skin lesions, but this process is limited by long time horizons and a multiplicity of interpretations. Deep learning techniques have been created to help dermatologists with these issues as a higher patient survival rate depends on the early and precise detection of skin cancer. This research proposed a new approach for binary classification of dermoscopic images for skin cancer. The Improved Grey Wolf Optimizer (I-GWO) is used in this technique to fine-tune some hyperparameters’ values of various pre-trained deep learning networks to maximize results. SqueezeNet, ShuffleNet, AlexNet, ResNet-18, and DarkNet-19 are the pre-trained networks that were employed. We tested the MED-NODE and DermIS databases in our investigation. Concerning the MED-NODE and DermIS datasets, the proposed method's highest accuracy results are 100% and 97%, respectively.

Uniqueness meets Semantics: A Novel Semantically Meaningful Bag-of-Words Approach for Matching Resumes to Job Profiles

Seba Susan, Muskan Sharma, Gargi Choudhary — Sun, 09 Jun 2024 00:00:00 +0200

In an increasingly competitive world, automated screening of resumes of candidates is the need of the hour given the large numbers of such resumes in career portals on the World Wide Web. Resume classification is a subset of the document classification problem in which the keywords extracted from the resume play a significant role in determining the job profile. In this paper, we explore the novel combination of uniqueness in terms of the number of occurrences of a keyword in a resume class as compared to the other resume classes, and the concept of semantics by representing the filtered keywords using word embeddings that can be used to find semantic similarities between resume documents. The principle of maximum entropy partitioning is used to find the keywords unique to a particular class. The aim is to use semantic representations of only those keywords that occur more frequently in one class more than in any other class; these are then passed as input to a Bidirectional long short-term memory (LSTM) for classification. Our experiments on a benchmark dataset proves that the proposed approach outperforms the state of the art in text classification by a significant margin proving the efficacy of our approach.

Leveraging Transfer Learning for Efficient Diagnosis of COPD Using CXR Images and Explainable AI Techniques

Victor Ikechukwu Agughasi — Wed, 12 Jun 2024 00:00:00 +0200

Chronic Obstructive Pulmonary Disease (COPD) is a predominant global health concern, ranking third in mortality rates, yet frequently remains undiagnosed until its advanced stages. Given its prevalence, the need for innovative and widely accessible diagnostic tools has never been more paramount. While spirometry tests serve as conventional diagnostic benchmarks, their reach remains limited, especially in regions with constrained medical resources. The presented research harnesses deep learning algorithms to facilitate early-stage COPD detection, specifically targeting Chest X-rays (CXRs). The clinically annotated VinDR-CXR dataset provides the primary foundation for model training, complemented by incorporating the ChestX-ray14 dataset for initial model pre-training. Such a dual-dataset strategy augments model generalization and adaptability. Among several explored Convolutional Neural Network (CNN) architectures, the Xception model emerges as a frontrunner. Through transfer learning methodologies, this model produces a noteworthy recall rate of 98.2%, markedly surpassing the metrics of the ResNet50 model. Recognizing the imperative for transparency in AI applications in medical imaging, the research integrates essential explainability approaches viz: Gradient Class Activation Mapping (Grad-CAM) and SHapley Additive exPlanations (SHAP). These techniques elucidate the AI’s decision-making process, offering invaluable visual and analytical insights for fostering trust among medical professionals. In essence, this study not only underscores the potential of integrating AI with medical imaging for COPD detection but also accentuates the pivotal role of transparency in AI-driven medical interventions.

DDoS Attacks Detection based on Machine Learning Algorithms in IoT Environments

Mehdi Ebady Manaa, Saba M. Hussain, Suad A. Alasadi, Hussein A. A. Al-Khamees — Thu, 11 Jul 2024 00:00:00 +0200

In today’s digital era, most electrical gadgets have become smart, and the great majority of them can connect to the internet. The Internet of Things (IoT) refers to a network comprised of interconnected items. Cloud-based IoT infrastructures are vulnerable to Distributed Denial of Service (DDoS) attacks. Despite the fact that these devices may be accessed from anywhere, they are vulnerable to assault and compromise. DDoS attacks pose a significant threat to network security and operational integrity. DDoS assault in which infected botnets of networks hit the victim’s PC from several systems across the internet, is one of the most popular. In this paper, three prominent datasets: UNSW-NB 15, UNSW-2018 IoT Botnet and recent Edge IIoT are using in an Anomaly-based Intrusion Detection system(AIDS) to detect and mitigate DDoS attacks. AIDS employ machine learning methods and Deep Learning (DL) for attack mitigation. The suggested work employed different types of machine learning and Deep Learning (DL): Random Forest (RF), Support Vector Machine (SVM), Logistic Regression, and Multi-layer perceptron (MLP), deep Artificial Neural Network (ANN), and Long Term Short Memory (LSTM) methods to identify DDoS attacks. Both of these methods are contrasted by the fact that the database stores the trained signatures. As a results, RF shows a promising performance with 100% accuracy and a minimum false positive on testing both datasets UNSW-NB 15 and UNSW-2018 Botnet. In addition, the results for a realistic Edge IIoT dataset show a good performance in accuracy for RF 98.79% and for deep learning LSTM with 99.36% in minimum time compared with other results for multi-class detection.

EpilConNet: A Novel Multi-Class Epileptic Seizure Classification Model

Sudip Ghosh, Deepti, Shivam Gupta — Wed, 31 Jul 2024 00:00:00 +0200

Epilepsy is a neurological disorder characterized by recurrent seizures, which can affect individuals of all age groups, but infants and older individuals are particularly vulnerable. Sudden epileptic attacks can pose significant risks and be life-threatening, impacting the overall quality of life of affected individuals. With the progress made in medical science, Electroencephalography (EEG) has emerged as a valuable tool for diagnosing and predicting seizure occurrences. The availability of wearable EEG devices, including caps and helmets, has become increasingly prominent in the market. As a result, there has been a recent surge in the development of deep learning-based systems. These systems are helpful for diagnosis in hospital settings and for mobile applications that provide timely warnings and predictions regarding seizure onset. Most of the existing state-of-the-art (SOTA) approaches focus on distinguishing between healthy and epileptic patients. Some studies categorize individuals into three classes: healthy, experiencing the onset of a seizure, or currently having a seizure, specifically focusing on mobile applications. However, limited literature is available on the five-class problem, which is valuable for localization and diagnosis in hospitals and mobile applications. In this regard, we propose our novel model, named EpilConNet, and conduct extensive experiments on a real-world dataset to demonstrate its efficacy in all modes of classification. EpilConNet results in a significant increase of 4% in accuracy in five-class classification.

Building an Integrative System: Enriching Gesture Language Video Recognition through the Multi-Stream of Hybrid and Improved Deep Learning Models with an Adaptive Decision System

Anwar Mira, Olaf Hellwich — Wed, 04 Sep 2024 00:00:00 +0200

The recognition of hand gestures is of growing importance in developing human-machine interfaces that rely on hand motions for communication. However, recognizing hand gesture motions poses challenges due to overlapping gestures from different categories that share similar hand poses. Temporal information has proven to be more effective in distinguishing sequences of hand gestures. To address these challenges, this research presents an innovative adaptive decision-making system that aim to enhance gesture recognition within the identical category have been introduced. The system capitalizes on the potential for variations in recognition outcomes derived from a diverse model of time-sharing neural networks, each employing different neural networks and trained on distinct input features. By incorporating such diverse input features, the system significantly boosts the robustness of recognition decisions, enabling it to effectively capture even the most subtle disparities within internal video representations. To achieve our research objective, we extensively investigate deep convolutional neural networks specifically trained on videos for hand gesture recognition. We also incorporate enhanced features from deep CNN using standard neural networks, namely Self Organizing Network and Radial Basis Function Network. By combining these features in various configurations, we develop novel frame-wise features based on the enhanced CNN features. These frame-wise features enable the training of diverse sets of recurrent neural network models, resulting in novel ensembles of composite models derived from various recurrent neural networks with diverse configurations. Some models are trained using multiple streams, while others utilize a single stream. To ensure the effective integration of these models, we implement a novel adaptive decision system mechanism that improves performance for weak prediction models and enhances overall recognition capability by taking a collective prediction decision. Experimental results demonstrate the significance of each proposed recurrent neural network model and the effectiveness of the new frame-wise features in enabling accurate decisions. This research achieves state-of-the-art performance in hand gesture recognition, highlighting the potential of combining different neural network architectures and feature representations to achieve superior outcomes.

Enhancing Noise Reduction with Bionic Wavelet and Adaptive Filtering

Shraddha C, Chayadevi M L, Anusuya M A, Vani H Y — Wed, 04 Sep 2024 00:00:00 +0200

Speech signals often contain different forms of background and environmental noise. For the development of an efficient speech recognition system, it is essential to preprocess noisy speech signals to reduce the impact of these disturbances. Notably, prior research has paid limited attention to pink and babble noises. This gap in knowledge inspired us to develop and implement hybrid algorithms tailored to handle these specific noise types. We introduce a hybrid method that combines the Bionic Wavelet transform with Adaptive Filtering to enhance signal strength. The performance of this method is assessed using various metrics, including Mean Squared Error, Signal-to-Noise Ratio, and Peak Signal-to-Noise Ratio. Notably, our findings indicate that SNR and PSNR metrics are especially effective in enhancing the handling of pink and babble noises.

AEPIA (Asociación Española para la Inteligencia Artificial). 40 Aniversario (1984 – 2024)

Fri, 17 May 2024 00:00:00 +0200

En este año se cumplen los primeros cuarenta años de la Asociación Española para la Inteligencia Artificial (AEPIA), cuarenta años de marcha continuada por el camino de la Inteligencia Artificial. Desde aquel lejano 1984, en el que un grupo de investigadores pioneros, liderados por el profesor D. José Cuena, fundador y primer presidente de AEPIA, vieron la necesidad de unir a toda la comunidad científica y profesional para dar a conocer la IA, hasta nuestros días, muchas generaciones han contribuido a la historia de AEPIA construyendo una asociación con muchos logros y muchos retos aún por delante.