International Conference on Applied Technologies

This research presents a recurrent neural network model that analyzes tweets and classifies them according to three categories: “positive”, “neutral” and “negative”, in order to identify supporting phrases in the semantic context of the topic under analysis. To do this, the Tweet Sentiment Extraction dataset and the LSTM neural model are used to train the network and evaluate, through network performance measurements and expert analysis, the precision and sensitivity of the model.

The results allow us to classify the sentiment behind a specific tweet with 86.78% accuracy, which shows that the network has a high level of precision, in relation to the expert’s assessment and comparing it with other works that have analyzed the sentiment. Dataset using Sentiment Analysis, Topic Modeling among other techniques. Additionally, a web dashboard is presented, which integrates the model analysis flow and allows the visualization of the results of the tweet classification.

In this way, the work offers a tool that can be used to classify the large amount of information on Twitter according to its polarity, recognizing and classifying patterns according to the semantic meaning of the terms. The applications of the model are many, for example, for understanding the characteristics of the personality of the public on social networks, the perception of customers in relation to products, services; which allows us to recognize opportunities to adjust marketing strategies, which benefit both clients and companies in general.

Medicine has evolved over time, utilizing technology to automate its relevant processes and expedite healthcare delivery. A notable example is electronic medical records, which would facilitate a doctor’s task when recording them, as it often becomes cumbersome and involves wasted time. However, many healthcare specialists with a higher age demographic find it challenging to adapt to digital or virtual environments for the completion of these medical records, which would streamline their documentation. Therefore, the focus of this present study is on a virtual assistant that, employing natural language processing, will serve the purpose of resolving the described problem, all through voice commands. The system was validated through a testing scenario with a sample of thirty users, using a success indicator that measures the percentage reduction in the time required to complete a medical history. Consequently, an average reduction of 54.1% in the time required for medical history recording was achieved. Additionally, 70% of the participants perceived that the virtual assistant would contribute to the optimization of medical history recording. Likewise, 80% considered the user interface to be pleasant and intuitive.

The present research was performed in the community of Luz de America, Santo Domingo de los Tsáchilas, Ecuador, during March-April (rainy season) and October (dry season) of 2022. The study assessed the quality of the drinking water consumed in the parish by collecting 176 samples in each season, as there have been no studies on water quality in the area. Physicochemical and microbiological parameters were analyzed according to the NTE INEN 1108 standard. The percentage of compliance with the standard was determined on the basis of the water source (well, user’s tap) and the municipalities under evaluation. The highest levels of non-compliance were found for residual chlorine and total coliforms in both seasons, while turbidity showed low compliance only in the rainy season. To determine the influence of seasonality, the Wilcoxon-Pratt Signed-Rank test was applied to the variables analyzed, all of which presented a p-value < 0.05, meaning that the medians of the difference between the two seasons for each variable were different, except for total coliforms (p-value = 0.092). Kendall’s correlation matrix was used to analyze the correlation between physicochemical and microbiological variables, showing 14 significant correlations in each climatic season, of which Turbidity-Colour was the highest for both rainy and dry seasons. Finally, a website was created for the stakeholders and the public, presenting the geo-referenced results obtained in the laboratory, ensuring compliance with the main water quality parameters for the benefit of the community.

This research is centered on the comprehensive analysis of meteorological data sourced from strategically positioned Global Navigation Satellite System (GNSS) stations located in Ecuador. Meteorological data of LJEC, PLEC, CUEC, and GZEC was collected and analyzed. For each station, three years (2017–2019) meteorological data recorded throughout each year at one-second intervals were analyzed. Data mining techniques are employed for in-depth analysis, utilizing machine learning algorithms to discern these stations behavior patterns. A machine learning model has been meticulously developed and fine-tuned to harmonize with the data’s underlying structure, thereby yielding precise results with a minimal margin of error. After model development and requisite testing, a satisfaction rate of 90% has been achieved, affirming formulated hypotheses validation.

Research aimed at obtaining physical measurements of animals in the wild generally makes use of chemical immobilizers to manipulate the object of study, which can be detrimental to the latter. This is why the present research of quantitative approach performs an experimental study that proposes the union of single-view metrology algorithms with the implementation of convolutional neural networks proposed in the YOLO model to develop a web application with two-layer architecture that can classify and take measurements of animals photographed with monocular camera traps in open spaces. This study returned positive results by allowing the development of a web page capable of taking measurements on 2D images with a margin of error of 0.55 cm in 0.49 s and classifying animals with an effectiveness of 93.85%, thus fulfilling the main objective of the study and contributing to the research gap.

The first step in preventing reductions in agricultural product output and quantity is to identify plant diseases. The research on plant diseases refers to examinations of patterns on the plant that may be observed with the naked eye. A vital component of sustainable agriculture is the observation of plant health and the identification of disease. The manual monitoring of plant diseases is highly challenging. It necessitates a huge amount of work, knowledge of plant diseases, and lengthy processing times. So, by taking photos of the leaves and comparing them to data sets, image processing is utilized to find plant illnesses. It is incredibly challenging to physically screen plant sicknesses. It requires a colossal measure of work, information on plant illnesses, and extended handling times. In this research, the diagnosis of rose plant diseases is critical for preventing yield and quantity losses in agricultural products. Plant disease identification is crucial for long-lasting agriculture. It requires a significant amount of work, a specialist understanding of plant diseases, and more than enough processing time. As a result, digital image processing is utilized to detect rose plant illnesses. Image acquisition, image pre-processing, picture segmentation, feature extraction, and classification are phases of disease detection. This research will look into how to save the rose plant from various diseases.

Parkinson’s Disease (PD) is a neurological condition that affects large masses of individuals around the world, having a great impact on the quality of life and raising notable challenges to the existing healthcare systems. Detection of PD at an early stage in an individual is crucial for the on-time intervention, diagnosis and improved patient outcomes. In the current scenario, various Machine Learning (ML) techniques, specifically, transfer learning-based approaches have proved to show promising results in analysis of medical images for diagnosis of diseases. A lot of work has been done in this area of utilizing transfer learning-based models such as VGG-19, ResNet etc. for the early detection of PD. Although most of these approaches have proved to have high performance values up to 95% accuracy, but there seems to have a large scope to improve upon and achieve higher performance measures that minimizes the risk of error in PD detection. This study uses the New Hand PD Dataset to propose an improved transfer learning model using fine-tuning specifically intended for the diagnosis of Parkinson's disease. The proposed novel approach in this paper uses feature extraction from the images and feeding them to fine-tuned transfer learning model, ResNet-152 leading to an improved testing accuracy of 100% and loss of 0.0040 only. To illustrate the suggested model's detection capability, its performance is compared with current cutting-edge deep learning and transfer learning models.

This paper develops a low-cost electronic visual assistant for visually impaired people, which determines the status of pedestrian traffic lights using artificial intelligence. By means of audio notifications, it describes the status of traffic lights, assisting blind people when moving around the streets. The computer vision and artificial intelligence software runs locally, avoiding the problem of latency in connectivity and is hosted on a Raspberry pi 4 development board. The transfer learning technique is used to obtain a reliable and adaptable convolutional neural network model for the problem of image identification. Employing conducting field tests on all the components of the prototype and its computer vision system, its effectiveness is verified. Through statistical analysis, detection results of 83.3% were obtained for red pedestrian traffic lights and an average of 77.7% in three out of the four categories for green traffic light detection. Additionally, a 100% error-free classification rate was determined, demonstrating its effectiveness and reliability.

When many medical images are obtained to complete a diagnostic test on a sample patient and the amount of radiation to which the human body is subjected. As a result, understanding how medical images are created is critical in the clinical setting. This region currently offers a wide range of options, which is convenient. For example, due to the unique clustering concept used by the fuzzy C-means (FCM) clustering methodology, the images produced by this method do not clearly indicate the attribution of specific firms. As a result, the image's finer details will become obscured, and the overall quality of the image will suffer. As a direct result of the GAN model's development, a plethora of novel techniques built on top of the deep generative adversarial network (GAN) model has emerged. Pix2Pix is based on the UNet model. This method employs two distinct medical photo types, as well as the calibration of a deep neural network, to generate high-quality photographs. There are strict data criteria, and the two sorts of medical pictures must be tailored to each individual patient. Transfer learning is used throughout the development of DualGAN models. The 3D image is divided into slices, and then simulations are run on each individual slice. The results of these simulations are then combined to produce the result. The disadvantage is that whenever a new image is created, “shadows” in the shape of bars present in the three-dimensional image will appear. Method or material. This study proposes a transfer learning based Dual3D & PatchGAN model as a solution to the problems described above and as a means of ensuring high-quality image production. Unlike traditional machine learning, which needs one-to-one matching of data sets, Dual3D and PatchGAN are based on transfer learning. As a result, just two distinct types of medical imaging data sets are required. This has a significant impact on the practical application of applications. By utilizing the images generated by DualGAN, this model can remove the bar-shaped “shadows” and transform the two different types of images in a different manner. Results: According to many evaluation indicators, Dual3D & PatchGAN are superior to other models in terms of their suitability for the development of medical images as well as their generation effect.

Before moving on with the structural parts of the research magnetic resonance imaging (MRI) scan is required because of its ability to highlight morphological changes in the brain over time, it has given researchers a unique viewpoint on the dynamic process by which the mind grows and adapts over one’s lifespan. As a result, they might have a more in-depth understanding of mental processes. The knowledge gathered in this manner has a monetary value that cannot be precisely represented. Because of the complexity of the data, the bulk of neuroimaging analytical pipelines rely on registration methods, which need labor-and time-intensive optimization processes. This is since mapping one image to another necessitates the use of registration procedures. The significance of registration in neuroimaging stems from the fact that this is the circumstance, which is why it exists. Recent deep learning algorithms have demonstrated the ability to accelerate the segmentation process. As an illustration, this is a risk because it raises the possibility of missing opportunities to precisely establish the boundaries of regions with uncertain borders. This is justified by the fact that giving up now would mean forfeiting the opportunity to accomplish anticipated future successes. This is especially crucial to remember when considering the challenge of multi-grained whole-brain segmentation, in which the size and structure of various parts of the brain might be highly diverse. We were able to make a deep learning network and use the information from this study to map the whole brain and figure out what its different parts are, this network can do so because it successfully partitions the brain. Given that its network can disassemble the brain into its constituent parts, this is a possibility. Our Multi-branch Residual Fusion Network (MRFNet) can quickly and precisely partition the whole brain into 136 subregions, making it far more efficient than the networks currently regarded to be the best in this field. The multi-branch cross-attention module (MCAM) allows for control over the organization’s more granular levels. As a result, the following actions were taken: Even if they know the full name, most individuals just use its abbreviation while discussing it. It has chosen that one of its primary goals will be to organize and synthesize the large amounts of granular context data that it will receive, and it has set a timeframe for achieving this goal. Another proposal is to use something called a residual error fusion module (REFM). We chose two separate datasets to demonstrate the validity and usefulness of the approach we developed for dissecting the whole brain. This will be done by comparing the results of the two datasets. The results show that the Proposed method is a reliable and effective way to find important parts of neuroimages before they are studied.

An artificial intelligence discipline has lately made considerable strides toward fully autonomous systems for classification and detection. Furthermore, with the aid of 5G networking and other next-generation wireless communications, the speed at which users may transport data while being undetectable to end users will increase. Because of a confluence of variables, the intelligent healthcare business is booming. There has never been a greater need for medical personnel to focus on the needs of their patients than today, in the aftermath of the COVID-19 epidemic. The pre-outbreak condition has changed dramatically. Vocal pathology accounts for a significant share of the population’s communication issues. If detected early enough, this illness is treatable and curable. This paper proposes a strategy for recognizing speech difficulties in the context of a hypothetical intelligent healthcare network. Devices used to capture speech activity, such as microphones and electroglottography (EGG) sensors, can be used to collect input for the Internet of Things (IoT). Before putting the input into a previously trained convolutional neural network, the signals are converted to spectrograms. The efficacy of the presented technique was proved by comparing the results to the publicly available Saarbrucken voice database. According to the experimental data, a bimodal input outperforms a single input in almost every regard. The recommended strategy has a 95.65% success rate, according to research.

The objective of this study was to experimentally examine the nature and significance of AI-infused leadership and its impact on subordinate employee job satisfaction in information technology projects. The present study employs an exploratory research design and utilizes a quantitative data methodology. An intricately designed questionnaire was devised to gather the data.To determine the sample size, the Cochran Formula is used, using a known population, a 95% confidence level, and a 5% margin of error. The calculated sample size is 100 respondents. Only entry and middle level employees from the top 10 IT companies in Bangalore city, based on market capitalization, are included. Specifically, we evaluate 10 individuals from each organization who have been involved in successful projects. The study's sample is obtained by easy sampling. The questionnaire's reliability and validity were assessed using the Gaskins’ master validity instrument, and it was confirmed. Data analysis in the study was conducted using SPSS Version 22 and AMOS Version 22 software. The incorporation of AI abilities in a leader's repertoire results in a significant 75% improvement in work satisfaction. Successful projects are characterized by leaders that provide ample AI-infused support to their subordinates, resulting in work satisfaction. Hence, enhancing the AI-integrated abilities of a leader is vital for the effective execution of initiatives. Moreover, the utilization of AI in leadership positions amplifies decision-making aptitude and streamlines procedures, hence augmenting efficiency and productivity. By using AI technology, leaders can effectively assess a larger volume of information and make decisions based on data, leading to improved results for both the team and the enterprise.

The virus primarily spread during immediate touch with contaminated people, and while researchers are still investigating other transmission, pathways physical touch has been considered a more likely mode. Traditional diagnosis methods had been become ineffective due to the rapid rise in infections. Researchers have created deep learning algorithms to deliver quick and precise COVID-19 diagnoses to solve the machine learning problem. The study comprises open COVID-19 datasets from various countries, and is separated into ML including the DL method. The paper provides a detailed description and comparison of the metrics used for evaluating the diagnosis procedures. For diagnosing COVID-19 and forecasting outbreaks, Convolution Neural Network is the considerably extensively utilized deep learning method, whereas the SVM approach is the most used ML algorithm. Future research in DL and ML policies toward COVID-19 diagnostics will be guided and inspired by this work.

Heart-related illnesses often known as CVDs (cardiovascular diseases) seem to be the leading cause of mortality globally in recent years. Consequently, a precise, workable, and trustworthy technique is necessary to recognize this disorder before time and begin the suitable treatment course. In this automated analysis of vast and complex health datasets, numerous machine learning methods are employed to scrutinize the information. Various machine learning techniques that have been developed by researchers are now being used by healthcare professionals to aid in the detection of heart-related disorders. Proposed study examines several models based on different methodological approaches, assessing the functionality of each. The Naive-Bayes model, SVM model (Support Vector Machines model), KNN model (K-Nearest Neighbor Model), DT model (Decision Trees Model), Ensemble models, and Supervised learning techniques based on RF model (Random Forest Model) are highly favored by researchers.

Growing knowledge of the organic diet as a healthy food alternative has spread around the world. The idea that organic food doesn't contain dangerous chemicals like nonorganic food has begun to gain widespread acceptance. More customers are turning to natural products as a result of the rising incidence of food contamination caused by the use of chemical pesticides and fertilisers in agriculture. More ecologically concerned consumers are gravitating toward organic foods. As a result, the organic food business worldwide is expanding at a pace of 20–22% yearly. Consumers’ lifestyles have evolved in recent years, and dining out has become commonplace. From a health perspective, the consumer is prepared to pay more for high-quality goods, particularly food products. New industry opportunities for producers and merchants have emerged as a result of promising customers a shift in mind set. Enhancing public health and protecting the environment both depend on organic agriculture. Therefore, encouraging organic farming will benefit customers as well as farmers by meeting consumer demands for high-quality food and environmentally friendly food production. This conceptual study lists 11 key elements that have a significant impact on customer intentions to buy organic goods. Further, this research investigates the impact of technology as a motivating factor behind consumers’ purchase intention for organic processed food. As the food industry experiences rapid advancements in technology, including digital platforms, e-commerce, and information-sharing tools, understanding how these technological elements influence consumer choices in the context of organic processed food becomes crucial.

Virtual assistants are used to complement user navigation and experience through e-commerce platforms and online services. Several studies show that customer experience is significantly improved when virtual assistants exhibit human-like attention and provide personalized recommendations tailored to individual preferences and needs. To achieve this, the assistant can make use of Natural Language Processing, which enables understanding of human language along with responses. This study develops a virtual assistant focused on conversational AI which is implemented in a poultry retail’s website to improve customer experience. To validate the benefits of the implementation, we collected feedback from a group of 58 customers in the city of Lima that interacted with the virtual assistant. In this way, results showed that the virtual assistant generated a positive impact on customer service with an average of 83.66% across dimensions such as usability, functionality, and customer satisfaction.

Occupational safety in construction sites is a vital topic due to the inherent risks associated with this type of work environment. This research provides a practical and effective solution to monitor compliance with safety standards. Therefore, it helps improve safety and minimize risks associated with not wearing safety helmets, such as head injuries and fatalities. The system is non-intrusive and can be easily integrated into existing surveillance systems. Moreover, it provides notifications to the administrator in case of non-compliance to enforce safety regulations. During the evaluation process, a set of video sequences recorded in real-world scenarios was used. The results obtained, mainly with the nano model, show that the system achieved a high level of precision with a score of 0.93, as well as an acceptable recall of 0.62. Regarding the F1 score, it has a score of 0.81.

Type 2 Diabetes Mellitus (T2DM) is one of the biggest threats to Ecuador’s health. The intake of processed foods has been linked to a higher risk of T2DM. This paper proposes FoodSub, a mobile application to recommend substitutes for processed foods using the NOVA Classification. Nutrient-based food clustering is used to identify substitute pairs between processed and unprocessed foods. The recommendations are supported and personalized using a knowledge graph that contains foods, dietary guidelines, and user information. In addition, a chatbot is implemented to answer simple questions about foods. This chatbot is developed using a Large Language Model (LLM) to query the knowledge graph. The mobile application and the chatbot are evaluated in terms of usability; both perform well, but there is room for improvement. Additionally, the recommendations’ performance is evaluated through expert verification. The recommendations perform well when issues like food transformation processes, flavor, context, or meal time are not relevant. Future work will consider the enhancement of the chatbot and the improvement of substitute recommendations for the relevant cases.

Software-defined networking represents a novel network model that separates control functionality from data management, significantly enhancing the latter’s efficiency and flexibility. Nevertheless, it faces substantial security threats that jeopardize data and service availability. This paper aims to define a model for classifying attacks using machine learning techniques to enhance defense capabilities and bolster data management security in software-defined networking. The classifier was trained with three machine learning algorithms: decision trees, random forests, and support vector machines, applying various feature sets from two public datasets with software-define networking traffic. In the training phase, 99.76%, 99.75%, and 99.50% accuracy rates were achieved for decision trees, random forests, and support vector machines, respectively. Consequently, the results obtained in this study outperform state-of-the-art approaches and demonstrate the successful deployment of a machine learning model in a software-defined networking environment.

This research addresses the problem of toxic comments in social networks, and how artificial intelligence (AI) and machine learning (Machine Learning) can help. It presents the development of a classification model using AI with machine learning techniques to identify toxic comments on Twitter.

The proposed classifier, developed in Python, was established with 7 different algorithms using approaches or strategies for multi-label classification, preprocessing, cleaning and data visualization. This model was trained with a total of 159571 comments from the Kaggle repository dataset called Jigsaw, which has the comments classified with various features. After the training, evaluation and comparison of the model created, the result was a classifier capable of identifying toxic and offensive words or comments with an accuracy of 92.16%.