Search

Potato crop has become integral part of our diet due to its wide use in variety of dishes, making it an important food crop. Its importance also stems from the fact that it is one of the cheapest vegetables available throughout the year. This makes it crucial to keep potato prices affordable for developing countries where the majority of the population falls under the middle-income bracket. Consequently, there is a need to develop a robust, effective, and portable technique to detect diseases in potato plant leaves. In this work, an attention-based disease detection technique is proposed. This technique selectively focuses on specific parts of an image which reveal the disease. This technique leverages transfer learning combined with two attention modules: the channel attention module and spatial attention module. By focusing on specific parts of the images, the proposed technique is able to achieve almost similar accuracy with significantly fewer parameters. The proposed technique has been validated using four pre-trained models: DenseNet169, XceptionNet, MobileNet, and VGG16. All of these models are able to achieve almost the same level of training and validation accuracy, around 90–97%, even after reducing the number of parameters by 40–50%. It shows that the proposed technique effectively reduces model complexity without compromising performance. [ABSTRACT FROM AUTHOR]

This research investigates the utilization of machine learning methodologies for the purpose of forecasting the fluctuations in stock values inside the financial market. The application of a Random Forest classifier is utilized on a dataset including historical stock prices (namely, the KSE-100 Index) to generate predictions regarding the future movement of stocks, specifically whether they would experience an increase or decrease. The model is trained via a sliding window methodology and is assessed through the utilization of precision, recall, and F1-score criteria. The study furthermore incorporates the utilization of back testing and hyper-parameter tweaking techniques in order to enhance the performance of the model. The findings indicate that the model demonstrates a precision score of 58%, representing an enhancement compared to the previous score of 48%. Nevertheless, the model's total accuracy stands at a mere 58%, underscoring the need for future enhancements. The report additionally proposes potential avenues for future research, such as exploring alternate data sources, employing sentiment analysis techniques, and developing more advanced algorithms. The findings of this study hold significant significance for investors and financial institutions, as they highlight the potential of machine learning in facilitating informed investment decisions and improving financial forecasts and analysis. [ABSTRACT FROM AUTHOR]

Background Machine vision faces significant challenges when applied to text recognition on cardboard packaging particularly due to multiple printing methods, irregular character shapes, and curved packaging surfaces. Methods This research introduces a novel deep learning application for recognizing binarized expiration date and batch code characters printed using multiple printing methods. The method, based on Region-based Convolutional Neural Networks (R-CNN), enables character recognition directly from in the images without the need for extracting handcrafted features. In detail, this approach performs character recognition by using the whole image as input, extracting and learning salient character features directly from the packaging surface images. Results The R-CNN model, with a precision of 91.1% and an F1 score of 80.9%, effectively recognizes manufacturing markings on pharmaceutical packages, with inconsistencies in the characters’ shapes. In a comparative experiment using the same dataset of images, the R-CNN model significantly outperformed Tesseract OCR, achieving much higher precision, recall, and F1 scores. Conclusions The results of this study reveal that the deep learning method outperforms the well-established optical character recognition method in recognizing text characters printed with different printing methods. Presented in this study, the deep learning method recognizes text characters with high precision. It is also suitable for recognizing text printed on curved surfaces, provided proper preprocessing is applied. The problem investigated in the study differs from previous research in the field, focusing on the recognition of texts printed with different printing methods. The research thus fills a gap in text recognition that existed in the research of the field. Furthermore, the study presents new ideas that will be utilized in our future research.

The creation of frameworks for lowering natural hazards is a sustainable development goal specified by the United Nations. This study aims to predict drought occurrence in Tamil Nadu, India, using 26 years of data, with only 3 drought years. Since the drought-occurrence years are minimal, it is an imbalanced dataset, which gives a suboptimal classification performance. The accuracy metric has a tendency to produce misleadingly high results by focusing on the accuracy of forecasting the majority class while ignoring the minority class; hence, this work considers the metrics' precision and recall. A novel strategy uses attribute (or instance) weighting, which allots weights to attributes (or instances) based on their importance, to improve precision and recall. These weights are found using a bio-inspired optimization algorithm, by designing its fitness function to improve precision and recall of the minority (drought) class. Since increasing precision and recall is a tug-of-war, multi-objective optimization helps to identify optimal attribute (or instance) weight balancing precision and recall while maximizing both. The newly introduced Synergistic Optimization Algorithm (SOA) is utilized for multi-objective optimization in order to ascertain weights for attributes (or instances). In SOA, to solve multi-objective optimization, each objective's population was generated using three distinct algorithms, namely, the Genetic, Firefly, and Particle Swarm Optimization (PSO) algorithms. The experimental results demonstrated that the prediction performance for the minority drought class was superior when utilizing instance (or attribute) weighting compared to the approach not employing attribute/instance weighting. The Gradient Boosting classifier with an attribute-weighted dataset achieved precision and recall values of 0.92 and 0.79, whereas, with instance weighting, the values were 0.9 and 0.76 for the drought class. The attribute weighting shows that in addition to the default drought indices SPI and SPEI, pollution factors and mean sea level rise are valuable indicators in drought prediction. From instance weighting, it is inferred that the instances of the months of March, April, July, and August contribute most to drought prediction. [ABSTRACT FROM AUTHOR]

The global dissemination of SARS-CoV-2 resulted in the emergence of several variants, including Alpha, Alpha + E484K, Beta, and Omicron. Our research integrated the study of eukaryotic translation factors and fundamental components in general protein synthesis with the analysis of SARS-CoV-2 variants and vaccination status. Utilizing statistical methods, we successfully differentiated between variants in infected individuals and, to a lesser extent, between vaccinated and non-vaccinated infected individuals, relying on the expression profiles of translation factors. Additionally, our investigation identified common causal relationships among the translation factors, shedding light on the interplay between SARS-CoV-2 variants and the host's translation machinery. [ABSTRACT FROM AUTHOR]

Deep Learning is becoming an emerging field in the Internet of Things (IoT) due to its ability to provide a comprehensive approach to automatic feature extraction and predictive modeling for analysis and decision-making. This paper introduces an IoT-based Dance Movement Recognition Model based on a Deep Learning Framework. The framework consists of a convolutional neural network (CNN) with a data-centric architecture to identify dance movements from the acquired data gathered by an IoT device. The IoT device collects 3D motion data captured by three accelerometers. Feature extraction is then done with the CNN architecture, resulting in a flattened matrix representing the movement. Subsequently, a Multi-Layer Perception (MLP) is used to classify the movements. The proposed system is experimentally evaluated on a standardized dataset of 16 dance steps with three-speed levels. The results show that our model outperforms state-of-the-art approaches in accuracy, evaluation time, and classification accuracy. The proposed model reached 90.74% accuracy, 87.12% precision, 83.78% recall and 84.39% F1-Score. The proposed model can serve as a basis for a reliable and intuitive system that can be used to monitor patient's dance movements with accuracy. [ABSTRACT FROM AUTHOR]

This research investigates the utilization of machine learning methodologies for the purpose of forecasting the fluctuations in stock values inside the financial market. The application of a Random Forest classifier is utilized on a dataset including historical stock prices (namely, the KSE-100 Index) to generate predictions regarding the future movement of stocks, specifically whether they would experience an increase or decrease. The model is trained via a sliding window methodology and is assessed through the utilization of precision, recall, and F1-score criteria. The study furthermore incorporates the utilization of back testing and hyper-parameter tweaking techniques in order to enhance the performance of the model. The findings indicate that the model demonstrates a precision score of 58%, representing an enhancement compared to the previous score of 48%. Nevertheless, the model's total accuracy stands at a mere 58%, underscoring the need for future enhancements. The report additionally proposes potential avenues for future research, such as exploring alternate data sources, employing sentiment analysis techniques, and developing more advanced algorithms. The findings of this study hold significant significance for investors and financial institutions, as they highlight the potential of machine learning in facilitating informed investment decisions and improving financial forecasts and analysis.

In recent years, the widespread adaptation of Unmanned Aerial Vehicles (UAVs), commonly known as drones, among the public has led to significant security concerns, prompting intense research into drones’ classification methodologies. The swift and accurate classification of drones poses a considerable challenge due to their diminutive size and rapid movements. To address this challenge, this paper introduces (i) a novel drone classification approach utilizing deep convolution and deep transfer learning techniques. The model incorporates bypass connections and Leaky ReLU activation functions to mitigate the ‘vanishing gradient problem’ and the ‘dying ReLU problem’, respectively, associated with deep networks and is trained on a diverse dataset. This study employs (ii) a custom dataset comprising both audio and visual data of drones as well as analogous objects like an airplane, birds, a helicopter, etc., to enhance classification accuracy. The integration of audio–visual information facilitates more precise drone classification. Furthermore, (iii) a new Finite Impulse Response (FIR) low-pass filter is proposed to convert audio signals into spectrogram images, reducing susceptibility to noise and interference. The proposed model signifies a transformative advancement in convolutional neural networks’ design, illustrating the compatibility of efficacy and efficiency without compromising on complexity and learnable properties. A notable performance was demonstrated by the proposed model, with an accuracy of 100% achieved on the test images using only four million learnable parameters. In contrast, the Resnet50 and Inception-V3 models exhibit 90% accuracy each on the same test set, despite the employment of 23.50 million and 21.80 million learnable parameters, respectively.

Data is essential for an organization to develop and make decisions efficiently and effectively. Machine learning classification algorithms are used to categorize observations into classes. The Naive Bayes (NB) classifier is a classification algorithm based on the Bayes theorem and the assumption that all predictors are independent of one another. Since this algorithm is based on probabilities, it is necessary to explore the sample distribution and feature type. This study presents an NB classifier method with enhanced performance among multidimensional and multivariate datasets, named the Naive Bayes Enrichment Method (NBEM). The NBEM is based on automated feature selection using threshold learning and the division of a dataset into sub-datasets according to the feature type. The main advantage of this method is the use of multiple NB classifiers based on different distributions and their combinations to classify a new observation. The final phase includes a weighted classification function that combines the results into a single output. This method was tested with 20 multivariate datasets and compared to other classification models and NB classifier variations. The results showed up to 76.3% improvement in the recall measure using NBEM and up to 43.9% improvement in the F1 score. Furthermore, we found that the error percentage of our method depended on the number of classes. [ABSTRACT FROM AUTHOR]

Predicting students' academic performance is of paramount importance to educational institutions. If students' academic performance is predicted course-wise, semester-wise and year-wise then it will be helpful for students to a great extent. In this research article, we generate twenty nine dataset with the help of students' result analysis. Datasets 1 to 9 is the datasets of students' result in first attempt while datasets 10 to 19 represent datasets of students after passing all courses of semesters and datasets 20 to 29 is overall dataset of students' result analysis. To predict the course-wise, semester-wise and year-wise performance of students, we developed a framework titled as E CSY APPS (E ducational C ourse S emester Y ear-wise A cademic P erformance Prediction S ystem) based on classification techniques and designed algorithm for analyzing students' performance in education sector. This E CSY APPS predicts the course-wise, semester-wise and year-wise grade of students. Fifteen classification algorithms such as Logistic Regression (LR), Naive Bayes (NB), Decision Tree (DT), Random Forest (RF), Support Vector Machines (SVM), Linear Support Vector Classification (LSVC), K-Nearest Neighbors (KNN), Gradient Boosting (GB), Adaptive Boosting (AdaBoost), Bagging, Extreme Gradient Boosting (XGBoost), Light Gradient-Boosting Machine (LightGBM), Categorical Boosting (CatBoost), Linear Discriminant Analysis (LDA), and Stochastic Gradient Descent Classifier (SGDC) are selected and applied on 29 datasets and compared on basis of performance parameters such as Accuracy, Precision, Recall, F1-score and Mean Absolute Error (MAE). These fifteen classification algorithms are under ten classifiers such as Linear Model, Naive Bayes, Tree, Support Vector Machine, Nearest Neighbors, Ensemble, xgboost, lightgbm, Discriminant Analysis and catboost. If the accuracy of two more classification algorithms for a dataset is same then in that case Precision, Recall, F1-score and Mean Absolute Error (MAE) are compared for deciding the best classification algorithm for the dataset. In this way, best classification algorithm is selected for each of 29 datasets. It is found that classifiers xgboost works best for eleven datasets while ensemble techniques Gradient Boosting and AdaBoost work best for two datasets and six datasets respectively among 29 datasets. Other classification algorithms such as Decision Tree, LightGBM, LDA, and KNN are noted to be best classification algorithm for two, four, three, and two datasets respectively. This framework is tested on new eight datasets related to students' result two methods such as K-Fold Cross-validation and Train-Validation-Test method. The results of this framework on new datasets shows that accuracy obtained on test dataset or validation dataset as compared to the accuracy obtained on old dataset is less than 6%. This framework will be helpful for students as well instructor. For students, it will help them to improve the performance of difficult courses from students'point of view in the examination while faculty can use this framework to improve pedagogical practices. [ABSTRACT FROM AUTHOR]