High Precision Values Research Articles

Artificial Intelligence System for Detecting the Use of Personal Protective Equipment

In recent years, occupational accidents have been increasing, and it has been suggested that this increase is related to poor or no supervision of personal protective equipment (PPE) use. This study proposes developing a system capable of identifying the use of PPEs using artificial intelligence through a neural network called YOLO. The results obtained from the development of the system suggest that automatic recognition of PPEs using artificial intelligence is possible with high precision. The recognition of gloves is the only critical object that can give false positives, but it can be addressed with a redundant system that performs two or more consecutive recognitions. This study also involved the preparation of a custom dataset for training the YOLO neural network. The dataset includes images of workers wearing different types of PPEs, such as helmets, gloves, and safety shoes. The system was trained using this dataset and achieved a precision of 98.13% and a recall of 86.78%. The high precision and recall values indicate that the system can accurately identify the use of PPEs in real-world scenarios, which can help prevent occupational accidents and ensure worker safety.

Validation of a Scale to Assess Activities of Daily Living at Home in Children and Adolescents With Autism Spectrum Disorder

Daily living skills may affect repetitive behavior patterns in individuals with autism spectrum disorder (ASD). This study aimed to validate a scale to assess activities of daily living at home (SAADL) in children and adolescents with ASD. Content validity by expert judgment showed Aiken V values between 0.75 and 0.88. The exploratory factor analysis (EFA) showed factor loadings (0.62 to 0.89) and high communalities. The % explanation of variance for dimension 1 (personal care) was 51.5% and for dimension 2 (mobility in the home) was 66.3%. The Cronbach's reliability was 0.87 and the retest values reflected high values of precision and accuracy. The SAADL is valid and reliable and can be used to assess activities of daily living of self-care and mobility at home.

Performance Evaluation of Different Object Detection Models for the Segmentation of Optical Cups and Discs.

Glaucoma is an eye disease that gradually deteriorates vision. Much research focuses on extracting information from the optic disc and optic cup, the structure used for measuring the cup-to-disc ratio. These structures are commonly segmented with deeplearning techniques, primarily using Encoder-Decoder models, which are hard to train and time-consuming. Object detection models using convolutional neural networks can extract features from fundus retinal images with good precision. However, the superiority of one model over another for a specific task is still being determined. The main goal of our approach is to compare object detection model performance to automate segment cups and discs on fundus images. This study brings the novelty of seeing the behavior of different object detection models in the detection and segmentation of the disc and the optical cup (Mask R-CNN, MS R-CNN, CARAFE, Cascade Mask R-CNN, GCNet, SOLO, Point_Rend), evaluated on Retinal Fundus Images for Glaucoma Analysis (REFUGE), and G1020 datasets. Reported metrics were Average Precision (AP), F1-score, IoU, and AUCPR. Several models achieved the highest AP with a perfect 1.000 when the threshold for IoU was set up at 0.50 on REFUGE, and the lowest was Cascade Mask R-CNN with an AP of 0.997. On the G1020 dataset, the best model was Point_Rend with an AP of 0.956, and the worst was SOLO with 0.906. It was concluded that the methods reviewed achieved excellent performance with high precision and recall values, showing efficiency and effectiveness. The problem of how many images are needed was addressed with an initial value of 100, with excellent results. Data augmentation, multi-scale handling, and anchor box size brought improvements. The capability to translate knowledge from one database to another shows promising results too.

Multispecies Machine Learning Predictions of In Vitro Intrinsic Clearance with Uncertainty Quantification Analyses.

In pharmaceutical research, compounds are optimized for metabolic stability to avoid a too fast elimination of the drug. Intrinsic clearance (CLint) measured in liver microsomes or hepatocytes is an important parameter during lead optimization. In this work, machine learning models were developed to relate the compound structure to microsomal metabolic stability and predict CLint for new compounds. A multitask (MT) learning architecture was introduced to model the CLint of six species simultaneously, giving as a result a multispecies machine learning model. MT graph neural network (MT-GNN) regression was identified as the top-performing method, and an ensemble of 10 MT-GNN models was evaluated prospectively. Geometric mean fold errors were consistently smaller than 2-fold. Moreover, high precision values were obtained in the prediction of "high" (>300 μL/min/mg) and "low" (<100 μL/min/mg) CLint compounds. Precision values ranged from 80 to 94% for low CLint predictions and from 75 to 97% for high CLint predictions, depending on the species. Uncertainty on experimental values and model predictions was systematically quantified. Experimental variability (aleatoric uncertainty) of all historical Novartis in vitro clearance experiments was analyzed. Interestingly, MT-GNN models' performance approached assays' experimental variability. Moreover, uncertainty estimation in predictions (epistemic uncertainty) enabled identifying predictions associated with lower and higher error. Taken together, our manuscript combines a multispecies deep learning model and large-scale uncertainty analyses to improve CLint predictions and facilitate early informed decisions for compound prioritization.

Geographical spatial analysis and risk prediction based on machine learning for maritime traffic accidents: A case study of Fujian sea area

Safety analysis according to the spatial distribution characteristics of maritime traffic accidents is critical to maritime traffic safety management. An accident analysis framework based on the geographic information system (GIS) is proposed to characterize the spatial distribution of maritime traffic accidents occurring in the Fujian sea area in 2007–2020 by employing kernel density estimation and spatial autocorrelation techniques. The sea area is divided into various grids, and in each grid, the mapping relationships between the number and severity of the traffic accidents and the traffic characteristics are established. Machine learning (ML) technology is used to assess whether a grid area is an accident-prone area and to predict accident severity in each grid. The accident prediction of different ML models, including random forest (RF) model, Adaboost model, gradient boosting decision tree (GBDT) model, and Stacking combined model, were compared. The optimality of the Stacking combined model was verified by comparing the experimental results of this model with those of classical prediction models, convolutional neural network (CNN), long short term memory (LSTM), and support vector machine (SVM). According to the results, the maritime accident data set of the entire Fujian sea area shows typical clustering characteristics and positive spatial correlation. That is, the kernel density estimation indicates that subareas, including the Ningde sea area, Fuzhou sea area, and Xiamen sea area, generally have high densities of maritime accidents and the highest risk value within the whole Fujian sea area. High-high accident clustering, that is high cluster areas neighbored by other areas of high cluster, is mainly seen in the Ningde and Fuzhou sea areas, while the Xiamen, Putian, and Zhangzhou subareas show low-low clustering, which are low clusters neighbored by low clusters. Among the ML models, the Stacking combined model shows high accuracy, precision, recall, and F1-score values of 0.912, 0.910, 0.912, and 0.904 in predicting whether a grid area is an accident-prone area and 0.750, 0.745, 0.750, and 0.746 in predicting the accident severity in the grid, indicating its superior maritime traffic accident prediction performance. Based on our analysis of the distribution characteristics and geospatial data, our proposed method demonstrates effective and reliable risk prediction.

Improving online education through automatic learning style identification using a multi-step architecture with ant colony system and artificial neural networks

Learning style is one of the individual differences which play an important role in learning. Being aware of it helps the student to understand their strengths and weaknesses, and the teacher to provide more valuable personalized interventions. Furthermore, learning style-based adaptive educational systems can be designed, which have been shown to increase student satisfaction or learning gain, while reducing the time needed to learn. It is therefore important to have an accurate method for identifying students’ learning styles. Since the traditional approach of filling in dedicated psychological questionnaires has several disadvantages, automatic methods have been proposed, based on investigating student observable behavior in a learning environment. Research done so far generally takes a mono-algorithmic approach to identify learning styles, and the precision rates leave room for improvement. Hence, in this paper we propose a novel hybrid multi-step architecture based on ant colony system and artificial neural networks to increase the precision of learning styles identification. Two different variants are proposed and evaluated with data from 75 students; results show high precision values, outperforming existing automatic approaches for learning style identification. The proposed architecture can be integrated into widely used educational systems (e.g., learning management systems) to provide learners and/or teachers with information about students’ learning styles. In addition, it can be integrated into adaptive educational systems and plugins of learning management systems to automatically identify learning styles and personalize instruction respectively.

In pursuit of the epileptogenic zone in focal epilepsy:a dynamical network biomarker approach

The success of resection surgery for drug-resistant epilepsy patients hinges on the correct identification of the epileptogenic zone (EZ) consisting of the subnetwork of brain regions that underlies seizure genesis in focal epilepsy. The dynamical network biomarker (DNB) method is a dynamical systems-based network analysis approach for identifying subnetworks that are the first to exhibit the transition as a complex system undergoes a bifurcation. The approach was devised and validated in the context of complex disease onset where the dynamics is known to be nonlinear and high-dimensional. We here adapt and implement the DNB approach for the identification of the EZ from the analysis of Stereoelectroencephalography (SEEG) data. The method is first successfully tested on simulated data generated with a large-scale brain network model of epilepsy using The Virtual Brain neuroinformatic platform and then applied to clinical SEEG data from focal epilepsy patients. The results are compared with those obtained by expert clinicians that designate the EZ using the Epileptogenicity Index (EI) method. High average precision values are obtained and posit the presented approach as a promising candidate tool for the pursuit of EZ in focal epilepsy.

Semantic Hashing for Fast Solar Magnetogram Retrieval

Abstract We propose a method for content-based retrieving solar magnetograms. We use the SDO Helioseismic and Magnetic Imager output collected with SunPy PyTorch libraries. We create a mathematical representation of the magnetic field regions of the Sun in the form of a vector. Thanks to this solution we can compare short vectors instead of comparing full-disk images. In order to decrease the retrieval time, we used a fully-connected autoencoder, which reduced the 256-element descriptor to a 32-element semantic hash. The performed experiments and comparisons proved the efficiency of the proposed approach. Our approach has the highest precision value in comparison with other state-of-the-art methods. The presented method can be used not only for solar image retrieval but also for classification tasks.

Using photographic records to quantify accuracy of bird identifications in citizen science data

Citizen science data are increasingly used for biodiversity monitoring. However, concerns are often raised over the accuracy of species identifications in citizen science databases, as data are collected mostly by non‐professionals. Misidentifications can simultaneously generate two error types: false positives (erroneous reports of a species) and false negatives (lack of reports of the misidentified species). Large‐scale assessments of identification errors should provide insights into the strengths and weaknesses of citizen science data. Here we show that citizen science photographic data for birds are trustworthy overall, although problems arise in hard‐to‐identify bird groups. We reviewed over 104 000 images of 377 passerine species from the southern Neotropics (Argentina) stored in eBird – a large citizen science platform – and quantified erroneous reports to calculate precision and recall metrics as measures for data accuracy. Precision increases with fewer false positives and recall increases with fewer false negatives; hence, high values of precision and recall will mirror a higher data accuracy. We found that 97% of the photos of all species were correctly identified. Most species (77%; n = 291) showed high accuracy in their identifications (precision and recall &gt; 95%), with 122 species showing no errors. A few hard‐to‐identify species (10%; n = 40) showed low levels of data quality (63–90% precision or recall). Similarly, few species (12%; n = 46) exhibited intermediate precision or recall scores (90–95%). Further, we uncovered the existence of a complex network of cross‐identifications composed of 272 species, with a predominance of tyrant flycatchers and ovenbirds, reflecting the strong traffic of errors that occurs within these families. To our knowledge, our study provides the first large‐scale quantification of identification errors in photos submitted by citizen science contributors. We underscore the relevance of performing such assessments to understand how identification errors are distributed across a database before analysing data, and provide tools for citizen science stakeholders to direct more specific efforts toward species that need an improvement in data quality.

INVESTIGATION OF THEENERGY DEPENDENCE OF THE INTERACTION POTENTIALS OF THE 16O+12С NUCLEAR SYSTEM WITH A SEMI-MICROSCOPIC METHOD

The study of the collision of heavy ions with light nuclei at low energy is important in nuclear physics, thermonuclear energy,and astrophysics. The high-precision values of the nuclear system described at an energy close to the Coulomb barrier are used to control the synthesis of light nuclei inside thermosynthesis. For cross sections of reactions of light nuclei on the sun, plasma,and stars, we can use the parameters we have set. The article presents a microscopic approach to describing the process of nuclear-nuclear interaction.In the phenomenological approach, which determines empirical values based on comparison with experimental data, it is possible to find many sets of parameters with potentials that characterize the cross-section well. But the question arises which of them are real.Therefore, it is necessary to additionally describe microscopic potentials. For the same reason, a semi-microscopic analysis was carried out, which describes the imaginary part of the nuclear potential on the basis of an optical model, and the real part on a double-folding model.The folding potential is constructed depending on the effective nucleon-nucleon interaction based on the matrix element of two nucleons and the density of the nucleon distribution.As a result of the analysis, differential cross sections and optimal parameters were determined that well characterize the experimental cross sections of the 16О+12C nuclear system at energies ELab=20, 24, 36 MeV. The coefficients of normalization of differential cross sections, described on the basis of real microfolding potentials, were determined in the range N=0.85-1.0.

Komparasi Fungsi Aktivasi Relu Dan Tanh Pada Multilayer Perceptron

Neural network is a popular method used in machine research, and activation functions, especially ReLu and Tanh, have a very important function in neural networks, to minimize the error value between the output layer and the target class. With variations in the number of hidden layers, as well as the number of neurons in each different hidden layer, this study analyzes 8 models to classify the Titanic's Survivor dataset. The result is that the ReLu function has a better performance than the Tanh function, seen from the average value of accuracy and precision which is higher than the Tanh activation function. The addition of the number of hidden layers has no effect on increasing the performance of the classification results, it can be seen from the decrease in the average accuracy and precision of the models that use 3 hidden layers and models that use 4 hidden layers. The highest accuracy value was obtained in the model using the ReLu activation function with 4 hidden layers and 50 neurons in each hidden layer, while the highest precision value was obtained in the model using the ReLu activation function with 4 hidden layers and 100 neurons in each hidden layer

Development Of Iris Based Age And Gender Detection System

Biometric recognition has shown their utility in different aspect including reducing search space significantly which has led to improve recognition performance, reduced computation time and faster processing time, a person is identified automatically by processing the unique features that are posed by the individual. The existing biometric systems recognize and accept a person, else simply reject a person if not enrolled, but these systems do not give software attributes of a person which is needed to search impostor as it has been observed in many occasions and especially in Nigeria that identification of an individual’s age and gender does not go beyond physical factors which in most cases can be manipulated or spoofed to carry out nefarious activities. This project focuses on developing an iris-based age and gender detection system for certain individuals to identify a person in real time. Left and right irises of 190 subjects constituting 1,716 images were acquired and Five Hundred and Seventy (570) left Irises were normalized to a uniform size of 250 by 250 pixels. Three Hundred and Forty-Two (342) images were used for training while the remaining Two Hundred and Twenty-Eight (228) were used for testing. The acquired images were preprocessed by performing segmentation, filtering and normalization using Daugman’s Rubber Sheet Model. Deep learning pre-trained networks are adopted to extract features from iris images. Further, these features are trained and classified using the multi-class Support Vector Machines (SVM) model for performance evaluation of the system. The system was implemented in Matrix Laboratory 9.0 (R2016a). The performance of the system was evaluated using accuracy, precision, recall and False Positive Rate (FPR). The hypothesis stating ‘the iris has age, and gender-related information’ is proven correct from experimental results. The evaluation results showed that False Positive Rate, Recall, Precision and Accuracy of gender prediction were 2.19%, 92.31%, 96.55% and 95.62% respectively at 0.75 threshold while for age prediction, the best values were obtained at age group of 20-24 for FPR, Recall, Precision and Accuracy which were 5.61%, 94.37%, 96.18% and 94.38% respectively. The developed model gave good performance with high recognition accuracy, recall, precision and low FPR values. Therefore, the developed model can be used in all firms and industries where security and personal identification is desired for security purpose and in going through investigation of criminal records for detecting age and gender of individuals.

Process Mining using Inductive Miner Algorithm to Determine the actual Business Process Model

At the beginning of 2019, the COVID-19 pandemic entered the country of Indonesia resulting in all learning activities being carried out online in all cities of Indonesia. Likewise, Telkom University concentrates all teaching and learning activities online using the CeLOE Learning Management System. Learning Management System is a system that helps lecturers in managing teaching and learning activities independently in educational institutions. CeLOE is a learning management system of Telkom University developed based on Moodle. In this study, we analyse the CeLOE event log using the process mining method. The goal is to find out the learning patterns of students using CeLOE during the COVID-19 pandemic. This research case study focuses on the activities of students of the Telkom University S1 Informatics study program for the first semester of 2020/2021 in using CeLOE LMS. The analysis of this study conducted a comparison of the performance of three variants of the inductive miner (IM) algorithm through conformance checking values. The results of the analysis obtained are value of conformance checking from the three variants of the inductive miner (IM) algorithm have an average fitness value of up to 1 prove that the inductive miner (IM) algorithm can make a model based on the event log well. Besides that, it has a fairly high precision value with a value range of 0.750-0.850 shows that the inductive miner (IM) makes a process model with relatively many variations of activities outside the event log and the IM process model is "overfit-ting" for all variants of the IM algorithm. Inductive miner (IM) is the best inductive miner (IM) algorithm variant with a fitness value of 1.0, precision value of 0.750, and the generalization value of this algorithm is relatively high (0.984). It is hoped that this research can contribute to the addition of new perspectives related to the implementation of process mining using inductive miner (IM) algorithm in the field of education

Performance Comparison Supervised Machine Learning Models to Predict Customer Transaction Through Social Media Ads

The application of machine learning has been used in various sectors, one of which is digital marketing. This research compares the performance of six machine learning algorithms to predict customer transaction decisions. The six algorithms used for comparison are Perceptron, Linear Regression, K-Nearest Neighbors, Naïve Bayes, Decision Tree, and Random Forest. The dataset is obtained from Facebook ads transaction data in 2020. The goal is to get a model that has the best performance so that it can be deployed to the web. The method that is used to compare the results is a confusion matrix and also uses visualization of the model to get the prediction error that occurred. Based on the test results, the random forest algorithm has the highest accuracy, recall, and f1-score values, with scores of 96.35%, 95.45%, and 93.32%. The highest precision value was generated by the logistic regression algorithm, which was 94.44%. Based on the data visualization presented by the random forest algorithm, it has the least prediction errors, there are four data. Therefore, it can be concluded that the random forest algorithm has the best performance because it has the highest value in the three confusion matrix measurements and the smallest data prediction error. The model of the random forest algorithm is deployed to the web platform and can be accessed at the link iklan-sosmed.herokuapp.com.

Electrochemical Immunoassay for Tumor Marker CA19-9 Detection Based on Self-Assembled Monolayer.

A CA19-9 electrochemical immunosensor was constructed using a hybrid self-assembled membrane modified with a gold electrode and applied to detect real samples. Hybrid self-assembled membranes were selected for electrode modification and used to detect antigens. First, the pretreated working electrodes were placed in a 3-mercaptopropionic acid (MPA)/β-mercaptoethanol (ME) mixture for 24 h for self-assembly. The electrodes were then placed in an EDC/NHS mixture for 1 h. Layer modification was performed by stepwise dropwise addition of CA19-9 antibody, BSA, and antigen. Differential pulse voltammetry was used to characterize this immunosensor preparation process. The assembled electrochemical immunosensor enables linear detection in the concentration range of 0.05–500 U/mL of CA19-9, and the detection limit was calculated as 0.01 U/mL. The results of the specificity measurement test showed that the signal change of the interfering substance was much lower than the response value of the detected antigen, indicating that the sensor has good specificity and strong anti-interference ability. The repeatability test results showed that the relative standard deviations were less than 5%, showing good accuracy and precision. The CA19-9 electrochemical immunosensor was used for the actual sample detection, and the experimental results of the standard serum addition method showed that the RSD values of the test concentrations were all less than 10%. The recoveries were 102.4–115.0%, indicating that the assay has high precision, good accuracy, and high potential application value.

Development and Validation of HPLC-UV Method for Determination of Paraquat in Raw and Commercial Samples

A new, rapid, efficient, low cost, robust, accurate and reproducible analytical method for the determination of Paraquat has been developed using reverse phase high performance liquid chromatography with UV-Visible detector (HPLC-UV). Paraquat is a frequently used herbicide. This method is useful for the detection and quantification of paraquat in raw materials and commercial samples with excellent recoveries upto almost 100%. pH of the mobile phase (acetonitrile:methanol:water 1:1:2) was optimized as 2.5 with ortho-phosphoric acid. This analytical method was validated with excellent linearity R2 = 0.999. LOD and LOQ were calculated to be 0.74 mg/L and 2.45 mg/L, respectively. The method showed a high precision (RSD%) value of 0.32, while the accuracy measured in terms of percentage recovery was almost 100% under optimized conditions. The robustness of the method was studied by changing the flow rate and the mobile phase concentration ratios, and the results obtained were within the permissible accepted values.

Performance Analysis for COVID-19 Diagnosis Using Custom and State-of-the-Art Deep Learning Models

The modern scientific world continuously endeavors to battle and devise solutions for newly arising pandemics. One such pandemic which has turned the world’s accustomed routine upside down is COVID-19: it has devastated the world economy and destroyed around 45 million lives, globally. Governments and scientists have been on the front line, striving towards the diagnosis and engineering of a vaccination for the said virus. COVID-19 can be diagnosed using artificial intelligence more accurately than traditional methods using chest X-rays. This research involves an evaluation of the performance of deep learning models for COVID-19 diagnosis using chest X-ray images from a dataset containing the largest number of COVID-19 images ever used in the literature, according to the best of the authors’ knowledge. The size of the utilized dataset is about 4.25 times the maximum COVID-19 chest X-ray image dataset used in the explored literature. Further, a CNN model was developed, named the Custom-Model in this study, for evaluation against, and comparison to, the state-of-the-art deep learning models. The intention was not to develop a new high-performing deep learning model, but rather to evaluate the performance of deep learning models on a larger COVID-19 chest X-ray image dataset. Moreover, Xception- and MobilNetV2- based models were also used for evaluation purposes. The criteria for evaluation were based on accuracy, precision, recall, F1 score, ROC curves, AUC, confusion matrix, and macro and weighted averages. Among the deployed models, Xception was the top performer in terms of precision and accuracy, while the MobileNetV2-based model could detect slightly more COVID-19 cases than Xception, and showed slightly fewer false negatives, while giving far more false positives than the other models. Also, the custom CNN model exceeds the MobileNetV2 model in terms of precision. The best accuracy, precision, recall, and F1 score out of these three models were 94.2%, 99%, 95%, and 97%, respectively, as shown by the Xception model. Finally, it was found that the overall accuracy in the current evaluation was curtailed by approximately 2% compared with the average accuracy of previous work on multi-class classification, while a very high precision value was observed, which is of high scientific value.

Radius of curvature measurement based on iterative correlation coefficient matching strategy with a single-shot interferogram

We proposed an iterative correlation coefficient matching strategy for the high-precision measurement of the radius of curvature (ROC), which only needs a single-frame interferogram. This strategy includes two parts: one is to obtain an initial ROC by using the relationship between the ROC and the diameter of the dark ring; other is to further solve a high-precision value by an iterative method. In the iterative part, we simulated a series of interference patterns with each ROC in the setting range, and calculated the correlation coefficient between the each simulated interferogram and the measured interferogram, and searched the maximum of the correlation coefficient that would be used to confirm a new ROC in the next iteration. When the calculated ROCs in the adjacent iterations meet the termination condition, the high-accuracy ROC can be solved. The feasibility of our proposed method has been verified by simulations and experiments. The errors of the test mirrors with the ROC of 101.6 mm and 1000 mm are 0.013% and 0.024% in experiments, respectively. This research provides a simple and feasible method for realizing the high-precision ROC measurement.

A comparative evaluation of convolutional neural networks, training image sizes, and deep learning optimizers for weed detection in alfalfa

AbstractIn this research, the deep-learning optimizers Adagrad, AdaDelta, Adaptive Moment Estimation (Adam), and Stochastic Gradient Descent (SGD) were applied to the deep convolutional neural networks AlexNet, GoogLeNet, VGGNet, and ResNet that were trained to recognize weeds among alfalfa using photographic images taken at 200×200, 400×400, 600×600, and 800×800 pixels. An increase in the image sizes reduced the classification accuracy of all neural networks. The neural networks that were trained with images of 200×200 pixels resulted in better classification accuracy than the other image sizes investigated here. The optimizers AlexNet and GoogLeNet trained with AdaDelta and SGD outperformed the Adagrad and Adam optimizers; VGGNet trained with AdaDelta outperformed Adagrad, Adam, and SGD; and ResNet trained with AdaDelta and Adagrad outperformed the Adam and SGD optimizers. When the neural networks were trained with the best-performing input image size (200×200 pixels) and the best-performing deep learning optimizer, VGGNet was the most effective neural network, with high precision and recall values (≥0.99) when validation and testing datasets were used. Alternatively, ResNet was the least effective neural network in its ability to classify images containing weeds. However, there was no difference among the different neural networks in their ability to differentiate between broadleaf and grass weeds. The neural networks discussed herein may be used for scouting weed infestations in alfalfa and further integrated into the machine vision subsystem of smart sprayers for site-specific weed control.

A Low Computational Cost Method for Mobile Malware Detection Using Transfer Learning and Familial Classification Using Topic Modelling

With the extensive use of Android applications, malware growth has been increasing drastically. The high popularity of Android devices has motivated malware developers to attack these devices. In recent times, most researchers and scholars have used deep learning approaches to detect Android malware. Although deep learning techniques provide good accuracy and efficiency, they require high computational cost to train huge and complex data sets. Hence, there is a need for an approach that can efficiently detect novel malware variants with a minimum computational cost. This paper proposes a novel framework for detecting and clustering Android malware using the transfer learning and the topic modelling approach. The transfer learning approach minimizes new training data by transferring well-known features from a qualified source model to a destination model, and hence, a high amount of computational power is not required. In addition, the proposed framework clusters the detected malware variants into their corresponding families with the help of Latent Dirichlet Allocation and hierarchical clustering techniques. For performance assessment, we performed several experiments with more than 50K Android application samples. In addition, we compared the performance of our framework with that of similar existing traditional machine learning and deep learning models. The proposed framework provides better accuracy of 98.3% during the classification stage by using the transfer learning approach as compared to other state-of-the-art Android malware detection techniques. The high precision value of 98.7% is obtained during the clustering stage while grouping the obtained malicious applications into their corresponding malware families.