Epoch Values Research Articles

EEGER – A Model for Recognition of Human Emotion Using Brain Signal

Emotions significantly impact human thinking, judgment, health, and communication. EEG-based emotion detection has advanced with the use of Brain-Computer Interface (BCI) technology, proving more effective than other physiological data. Despite progress in affective computing, emotion recognition remains a challenge. However, it's increasingly common in brain–machine interfaces, and research shows EEG brain waves are valuable in identifying emotional states. This research introduces a novel automated system for emotion recognition using deep learning techniques on EEG data collected from the GAMEEMO dataset, where participants played emotional assessment games. The system is designed to identify four emotions experienced during gameplay. The proposed model, called EEGER, was trained exclusively on EEG signals and demonstrated a 99.99% accuracy with minimal computational time. Key to its efficiency is the use of LSTM (Long Short-Term Memory) classifiers, which simplify the process by automatically extracting relevant features. The system was tested across different learning rates and epoch values, showing that 10 epochs with a learning rate of 0.0001 were sufficient to achieve the best accuracy. EEGER was also compared with other methods like K-Nearest Neighbor (KNN), Linear Discriminant Analysis (LDA), and Adaptive Boosting (AdaBoost), outperforming them in both accuracy and efficiency. These findings suggest that EEGER offers a promising new approach to EEG-based emotion recognition, optimizing performance with lower complexity and computation time.

Lung Cancer Detection Using a Modified Convolutional Neural Network (CNN)

<p><span>Image processing is used to classify lung images with malignant or normal nodules. The Convolutional Neural Network (CNN) method is often used to classify images. This study uses a modified CNN architecture with various layers, filters, batch size, dropout, and epoch values. Variations were made to determine the best accuracy value and reduce the overfitting value of the proposed CNN architecture. This study implements the method using the Keras library with the Python programming language. The data is in the form of CT-Scan images of lung cancer and normal lungs. The results of several experiments from the proposed model produce an accuracy value of 95% using three layers, 128 filters on the first layer, 256 on the second layer, and 512 filters on the third layer, then with 32 batch sizes, 0.5 dropout.</span></p>

Refining tomato disease recognition: hyperparameter tuning on ResNet-101 architecture for precise leaf-based classification

Tomatoes plants are widely recognized as versatile vegetables globally. This study aims to develop a high-precision web interface for classifying various leaf diseases in tomatoes. Utilizing a convolutional neural network (CNN) algorithm using the residual network-101 (ResNet-101) architecture, this tool aids farmers in accurately identifying leaf diseases in tomatoes, thereby reducing the risk of crop failure. The dataset comprises 6,800 images, categorized into four classes: early blight, spider mites two spotted, tomato yellow leaf curl virus, and healthy tomatoes, each containing 1,700 images. Hyperparameter tuning was conducted as part of an experiment aimed at enhancing the performance of the model. Experiments involved varying epoch values (10, 25, 30, 50, 60, 75, 100, and 110), a fixed batch size of 4, different learning rates (0.1, 0.01, 0.001, 0.0001), and num workers (4, 8, 16). The results demonstrated an accuracy of 99% with 100 epochs, a batch size of 4, a learning rate of 0.001, and 16 num workers. Consequently, this research contributes to a deeper understanding of disease management in tomato plants, ensuring optimal quality and quantity of the harvest.

An approach based on deep learning methods to detect the condition of solar panels in solar power plants

Solar panels are increasingly popular due to global energy shortages and rising costs. However, managing large or elevated panel systems requires regular oversight, leading to potential time and cost challenges. This study was focused on developing an AI-based drone for panel detection to address these issues and facilitate the control process. A low-cost system for AI-based identification of dusty, broken, and healthy solar panels was created using a Raspberry Pi 4B board and camera. The study proposed a Histogram Equalization (HE)-based preprocessing technique to improve the AI model. Firstly, the trainings were performed with YOLOv5 without the proposed method at epoch values of 100, 150, and 200 in order to see the effectiveness of the proposed method more clearly. As a result of these trainings, the highest F1 score was obtained as 80 %. In the second step, three deep learning algorithms - YOLOv5, YOLOv7, and YOLOv8 - with epochs of 100, 150, and 200 respectively, were used for training with the proposed method. A detailed comparative analysis of the developed models was carried out regarding their performance metrics. The YOLOv5l was obtained as the most successful panel detection model with an F1 score of 97 % at 150 epochs. The model with the best performance metrics was used in a real-time test application with an AI-based drone. F1 score results were obtained between 90 % and 97 %, mainly supporting the success rate obtained in real-time application. The results strongly support the effectiveness of this proposed method for panel detection tasks, showcasing its high efficacy and promising potential.

트랜스포머 모델을 활용한 직무 추천 모델 개발에 관한 연구

With the rapid technological advancements brought about by the Fourth Industrial Revolution, the industrial ecosystem is undergoing changes with new professions emerging and existing ones disappearing at an unprecedented rate. This societal shift is leading to an employment market dominated by professionals with experience, with an increasing number of individuals working in roles unrelated to their major. Additionally, the discrepancy between one's major and the job requirements demanded by industries is leading to a rising number of young job seekers abandoning their job search. In light of this, the purpose of this study is to develop job recommendation models based on job seekers' resumes, evaluate the performance of these models, and propose ways to enhance the applicability of the recommended jobs. To achieve this, two experiments were conducted. The first experiment evaluated the performance of a job recommendation model that applied the Transformer architecture, using a training dataset composed of resumes with 52 features and set hyperparameters. The second experiment evaluated the performance of a deep learning model, which applied fine-tuning techniques on a training dataset constructed from 34 features after removing any feature with more than 30% missing values from the original 52 features used in the first experiment. The increase in the epoch value of the Transformer model has significant implications.
 This research demonstrates that the optimization of training data and the application of fine-tuning in deep learning model design significantly impact the performance improvement of job recommendation models. Given the rate of increase in epoch values observed in this study, it is expected that applying high-quality resume data to the Transformer model for additional model training in the future will further enhance the performance of career-based job recommendation models for both students preparing for employment based on their major and for experienced job seekers.

Development of a Mobile Application for Plant Disease Detection using Parameter Optimization Method in Convolutional Neural Networks Algorithm

Plant diseases are a serious problem in agriculture that affects both the quantity and quality of the harvest. To address this issue, authors developed a mobile software capable of detecting diseases in plants by analyzing their leaves using a smartphone camera. This research used the Convolutional Neural Networks (CNN) method for this purpose. In the initial experiments, authors compared the performance of four deep learning architectures: VGG-19, Xception, ResNet-50, and InceptionV3. Based on the results of the experiments, authors decided to use the CNN Xception as it yielded good performance. However, the CNN algorithm does not attain its maximum potential when using default parameters. Hence, authors goal is to enhance its performance by implementing parameter optimization using the grid search algorithm to determine the optimal combination of learning rate and epoch values. The experimental results demonstrated that the implementation of parameter optimization in CNN significantly improved accuracy in potato plants from 96.3% to 97.9% and in maize plants from 87.6% to 93.4%.

Application of deep learning technique in next generation sequence experiments

In recent years, the widespread utilization of biological data processing technology has been driven by its cost-effectiveness. Consequently, next-generation sequencing (NGS) has become an integral component of biological research. NGS technologies enable the sequencing of billions of nucleotides in the entire genome, transcriptome, or specific target regions. This sequencing generates vast data matrices. Consequently, there is a growing demand for deep learning (DL) approaches, which employ multilayer artificial neural networks and systems capable of extracting meaningful information from these extensive data structures. In this study, the aim was to obtain optimized parameters and assess the prediction performance of deep learning and machine learning (ML) algorithms for binary classification in real and simulated whole genome data using a cloud-based system. The ART-simulated data and paired-end NGS (whole genome) data of Ch22, which includes ethnicity information, were evaluated using XGBoost, LightGBM, and DL algorithms. When the learning rate was set to 0.01 and 0.001, and the epoch values were updated to 500, 1000, and 2000 in the deep learning model for the ART simulated dataset, the median accuracy values of the ART models were as follows: 0.6320, 0.6800, and 0.7340 for epoch 0.01; and 0.6920, 0.7220, and 0.8020 for epoch 0.001, respectively. In comparison, the median accuracy values of the XGBoost and LightGBM models were 0.6990 and 0.6250 respectively. When the same process is repeated for Chr 22, the results are as follows: the median accuracy values of the DL models were 0.5290, 0.5420 and 0.5820 for epoch 0.01; and 0.5510, 0.5830 and 0.6040 for epoch 0.001, respectively. Additionally, the median accuracy values of the XGBoost and LightGBM models were 0.5760 and 0.5250, respectively. While the best classification estimates were obtained at 2000 epochs and a learning rate (LR) value of 0.001 for both real and simulated data, the XGBoost algorithm showed higher performance when the epoch value was 500 and the LR was 0.01. When dealing with class imbalance, the DL algorithm yielded similar and high Recall and Precision values. Conclusively, this study serves as a timely resource for genomic scientists, providing guidance on why, when, and how to effectively utilize deep learning/machine learning methods for the analysis of human genomic data.

Implementation of Convolutional Neural Network Algorithm Using Vgg-16 Architecture for Image Classification in Facial Images

Abstract: Face Recognition has now become a technology capable of recognizing facial patterns, facial image recognition is also used in various applications, for example in biological data recognition applications, digital image and video search, room security, and other applications.
 Purpose: This study aims to find out how the implementation of the CNN method with the VGG-16 architecture affects the classification of gender in facial images and how it affects the results.
 Methods/Study design/approach: In this study, we use the CNN method for data processing and build the program and use VGG-16 Architecture to build the model, then the tensorflow library for calling the required features such as when optimizing or for statistical plots and using the Confusion Matrix to determine the level of accuracy obtained. The desired results in this study are accuracy, precision, recall, and Fscore.
 Result/Findings: Classifying facial images using CNN with VGG-16 architecture provides an accuracy rate of 94%. From the results of this study it can be concluded that the model with the best accuracy is at epoch 20 compared to epoch 60, epoch 80, and epoch 100 which have previously been tested.
 Novelty/Originality/Value: The level of accuracy resulting from the implementation of the CNN method using the VGG-16 Architecture for image classification in facial images is quite good, resulting in an accuracy of 94%. Accuracy results were obtained from tests carried out by comparing several epoch values to produce the best accuracy of 94% using epoch 20.

Classification Appropriateness Recipient Help Non-Cash Food Using Learning Vector Quantization (LVQ) Method

Help Non-Cash Food is a program from the Government that is used to overcome poverty. The program is not functioning as well as it could because the procedure of receiving aid is not uniform, and individuals responsible for making choices are having trouble determining which families are qualified to receive the assistance. To overcome this problem, a classification system is needed to classify the eligibility of Non-Cash Food Assistance recipients so that the results are more efficient and accurate. This research uses the Learning Vector Quantization (LVQ) method with Python. This research aims to implement the LVQ method for the eligibility classification of non-cash food assistance recipients. System design is a stage that contains the process from start to finish of running this system which is described in the form of a flowchart, including system requirements that support this research, both software and hardware. In the process of analyzing the results and tests that are used as evaluation material in the process of finding a solution to a problem and making decisions in the process of planning activities, it is necessary to assess whether or not the LVQ approach is practicable to apply based on the findings of the research. In this study, 200 datasets were used with three epoch values and a learning rate of 0.1. The data set was randomly divided into a training portion of 80% and a testing portion of 20%. So that the results of this research using the LVQ method on the eligibility classification of recipients of Non-Cash Food Assistance obtain an accuracy of 97.5%.

URBAN MAP GENERATION IN ARTIST’S STYLE USING GENERATIVE ADVERSARIAL NETWORKS (GAN)

Artificial Intelligence is a field that is able to learn from existing data to synthesize new ones using deep learning methods. Using Artificial Neural Networks that process big datasets, complex tasks and challenges become easily resolved. As the zeitgeist suggests, it is possible to produce novel outcomes for future projections by applying various machine learning algorithms on the generated data sets. In that context, the focus of this research is exploring the reinterpretation of 21st century urban plans with familiar artist styles using different subtypes of deep-learning-based generative adversarial networks (GAN) algorithms. In order to explore the capabilities of urban map transformation with machine learning approaches, two different GAN algorithms which are cycleGAN and styleGAN have been applied on the two main data sets. First data set, the urban data set, contains 50 cities urban plans in .jpeg format collected according to the diversity of the urban morphologies. Whereas the second data set is composed of four well-known artist’s paintings, that belong to various artistic movements. As a result of training the same data sets with different GAN algorithms and epoch values were compared and evaluated. In this respect, the study not only investigates the reinterpretation of stylistic urban maps and shows the discoverability of new representation techniques, but also offers a comparison of the use of different image to image translation GAN algorithms.

Evaluating TensorFlow image classification in classifying proton collision images for particle colliders

The Large Hadron Collider at the European Organization for Nuclear Research (CERN) is planned to increase the number of proton collisions 10-fold by 2025. To keep up with this rapid pace, novel and more efficient particle collision classification methods must be developed, which many physicists believe can be achieved via machine learning. Since particle collisions and trajectories are mapped onto images, we chose TensorFlow’s image classification model to analyze and predict particle classes from these collision images. We aimed to evaluate the accuracy at which image classification could correctly identify a particle’s category from its trajectory image. We hypothesized that image classification would improve its accuracy with larger training datasets and more training epochs. To test our hypotheses, we randomly partitioned 100 images for testing and reserved 3,400 for training. We ran numerous trials using different training set sizes, different training epoch values, and binary classification. Our results show that the model’s accuracy increases with more training samples, and it increases its accuracy and consistency with more training epochs. In binary classification, the model distinguishes the electron class the best, with a 95.2% mean accuracy. We conclude that image classification has remarkable implementation potential in classifying particle collision outcomes because of its ability to improve with training sample size and in binary classification. Our work highlights one method collision classification algorithms should utilize, which can work effectively with the increased data output by particle colliders and give physicists a vital tool in accelerating their field.

Convolutional Neural Network – Long Short Term Memory Untuk Prediksi Harga Emas Indonesia

In the economic field, gold generally has three main functions, namely the monetary function, investment and industrial functions. In the financial world, prediction of the trend of gold price fluctuations is an important issue. Convolutional neural network algorithm is one of the popular algorithms in the image classification domain. But this algorithm can also be applied to 1-dimensional problems, such as predicting the next value in a sequence, be it a time series or the next word in a sentence. In this study, the Convolutional Neural Network algorithm was optimized with long shot time memory to predict the price of Indonesian gold from 1979 to 2021 with a total of 511 records. The test was carried out with three different epoch values, namely 50, 100 and 150 with a validity test using RMSE. From the test results, the smallest RMSE train value and RMSE validation are generated at epoch 150, which is 0.0138 for RMSE train and 0.0402 for RMSE validation.

Application of Machine Learning on Air Quality

Air pollution can cause serious health problems and millions of people live in areas with poor air quality. Understanding air quality is an important topic of research which in turn helps understand health and global climate. In this paper, I applied machine learning techniques to the predict the air quality. Using a data set collected in urban cities with information on air pollutants, supervised learning was applied to gain insights and predict quality. Neural networks are used to determine the specific factors that impact the air quality index and the accuracy of the prediction. The research was helpful in identifying specific factors that affected air quality and accuracy of prediction improved when increasing hidden layers and epoch values within the neural networks.

CATTLE DETECTION AND MISHAP AVOIDANCE SYSTEM USING YOLO v5 ALGORITHM

Aim: We propose a robust vehicular cattle detection model using YOLO version 5 for vehicles to avoid mishap on Indian roads. Results: The research was conducted with the help of training and validation on-road cattle image dataset and tested the model for various epoch values. Conclusion: The model has predicted 82% to 85% of true positive result with 90.5% accuracy and fruitful test results observed on real world video samples.

Radicalism Speech Detection in Indonesia on Twitter Using Backpropagation Neural Network Method

In this modern era, many people use social media easily and freely. One of the social media used is Twitter. The reason people use Twitter is that they can express their opinion freely. However, this freedom does not always have a positive impact on other Twitter users. One of the negative impacts for users is that they can spread radical content. Therefore, this research aims to detect whether a tweet contains radical elements or not using the backpropagation neural network method. The process is carried out by taking data on Twitter, after which the preprocessing process is carried out. Then the data is processed using imbalanced handling, where the data is divided into oversampling and undersampling data. After the data is divided, the next process is to do stopword and then look for accuracy by comparing different epoch values, namely 100, 150, 200, and 250. The best epoch value obtained is 200, with a final accuracy result of 86%.

Implementation of deep learning methods in prediction of adsorption processes

The article presents deep learning methods applied to predict the mass of an adsorption bed in the fixed and fluidized bed. The purpose of the application of this kind of bed is to improve the efficiency of the adsorption cooling systems by increased heat and mass transfer by using fluidization. The paper employs three deep learning methods: Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM), and Gated Recurrent Unit (GRU). In each of the selected neural networks, the epoch values, the number of neurons on the first layer and the number of neurons on the second layer were changed. These networks had two hidden layers. The paper presents numerical research on mass prediction using the algorithm mentioned above for silica gel as sorbent with copper, aluminum, carbon nanotubes additives. The results obtained by the developed algorithms of the LSTM, BiLSTM, GRU network and experimental tests are in good agreement with R2 above 0.97. The GRU network guarantees predicting the mass of both the fluidized and fixed beds with the best agreement with the measurement results.

Comparison of multi-class and fusion of multiple single-class SegNet model for mapping karst wetland vegetation using UAV images

Wetland vegetation classification using deep learning algorithm and unmanned aerial vehicle (UAV) images have attracted increased attentions. However, there exist several challenges in mapping karst wetland vegetation due to its fragmentation, intersection, and high heterogeneity of vegetation patches. This study proposed a novel approach to classify karst vegetation in Huixian National Wetland Park, the largest karst wetland in China by fusing single-class SegNet classification using the maximum probability algorithm. A new optimized post-classification algorithm was developed to eliminate the stitching traces caused by SegNet model prediction. This paper evaluated the effect of multi-class and fusion of multiple single-class SegNet models with different EPOCH values on mapping karst vegetation using UAV images. Finally, this paper carried out a comparison of classification accuracies between object-based Random Forest (RF) and fusion of single-class SegNet models. The specific conclusions of this paper include the followings: (1) fusion of four single-class SegNet models produced better classification for karst wetland vegetation than multi-class SegNet model, and achieved the highest overall accuracy of 87.34%; (2) the optimized post-classification algorithm improved classification accuracy of SegNet model by eliminating splicing traces; (3) classification performance of single-class SegNet model outperformed multi-class SegNet model, and improved classification accuracy (F1-Score) ranging from 10 to 25%; (4) Fusion of single-class SegNet models and object-based RF classifier both produced good classifications for karst wetland vegetation, and achieved over 87% overall accuracy.

Recognition of Regional Traditional House in Indonesia Using Convolutional Neural Network (CNN) Method

Indonesia is a country that has a lot of cultural diversity. This cultural diversity needs to be preserved. If this is not done, the culture that is owned by Indonesia can slowly disappear. The reduction in cultural values can also reduce the sense of belonging to the culture. This lack of sense of ownership makes it easy for other nations to make claims on the culture that is owned by Indonesia. Indonesia will lose its characteristics as a country that has a lot of cultural diversity. One of the efforts to preserve culture is to recognize the characteristics of each culture and be able to recognize the differences between one culture and another. For example, recognizing traditional houses from various ethnic groups based on their image. In this research, the image classification of the characteristics of traditional houses from several ethnic groups in Indonesia was carried out. The classification used to identify an image. In this study, deep learning techniques are used with the Convolutional Neural Network (CNN) algorithm and Keras framework. This CNN use several layers namely convolutional, pooling, flatten, and dense layer. The development of deep learning models uses the Knowledge Discovery in Database (KDD) method. This method consists of nine stages. The built model is evaluated using k-fold cross validation with a k value of 5 and produces an average accuracy of 80%. This shows that the model built is capable of classifying well. The built model is evaluated with 3 different epoch values, namely 50, 75, and 100. The larger the epoch value used, the greater the accuracy value. The model built is also able to make predictions with an accuracy of 80%.

Identification of Corn Leaf Diseases Comprising of Blight, Grey Spot and Rust Using DenseNet-201

Corn is a vital commodity in Malaysia because it is a key component of animal feed. The retention of the wholesome corn yield is essential to satisfy the rising demand. Like other plants, corn is susceptible to pathogens infection during the growing period. Manual observation of the diseases nevertheless takes time and requires a lot of work. The aim of this study was to propose an automatic approach to identify corn leaf diseases. The dataset used comprises of the images of diseased corn leaf comprising of blight, grey spot and rust as well as healthy corn leaf in YCbCr colour space representation. The DenseNet-201 algorithm was utilised in the proposed method of identifying corn leaf diseases. The training and validation analysis of distinctive epoch values of DenseNet-201 were also used to validate the proposed method, which resulted in significantly higher identification accuracy. DenseNet-201 succeeded 95.11% identification accuracy and it outperformed the prior identification methods such as ResNet-50, ResNet-101 and Bag of Features. The DenseNet-201 also has been validated to function as anticipated in identifying corn leaf diseases based on the algorithm validation assessment.

FOREST FIRE DETECTION USING CONVOLUTIONAL NEURAL NETWORK

Fire outbreaks are a typical occurrence all throughout the world, and the harm they bring to both nature and humans is enormous. In comparison to classic sensor-based fire detection systems, vision-based fire detection systems have recently gained favour. The detection process using image processing techniques, on the other hand, is extremely time consuming. We designed a Convolutional Neural Network-based fire detection technique. Through training with datasets, obtain high-accuracy fire image detection that is consistent with fire detection. In this paper, we propose a system for detecting fire using Convolutional Neural Networks (CNN). This paper critically analyzes the scope of Artificial intelligence for detection with video from CCTV footages. This project uses dataset containing video frames with fire. The data is then preprocessed and use the CNN to build a model to detect fire. The dataset is given as input for validating the algorithm and experiments are noted. This project focus on building high accuracy and cost efficient machine that can be used for fire detection.There are 755 and 244 images in the datasets for fire and non-fire, respectively. There are 999 photos in all. These images were created using fire-related video and some images found on the internet. There are 999 images that have been scaled and reshaped to convert to a training dataset and 999 images that have been resized and reshaped to convert to a testing dataset. Convolution, activation functions, and the max pooling technique are used to train the model. The model is trained by varying batch sizes and epoch values. As a result, we obtain a high accuracy and detection rate. Key words: CNN,fire detection,machine learning.