Holdout Method Research Articles

K-Nearest Neighbor and Feature Extraction on Detection of Pest and Diseases of Cocoa

Knowledge and utilization of digital images are growing rapidly not only in the fields of medicine and industry but also in the field of agriculture. This knowledge can apply it to a computer-based program that is used to detect agricultural products more effectively and efficiently. this research aims to build a system to detect the types of pests and diseases of cocoa pods because in general, an inspection of pests and diseases of cocoa pods is still manual based on the visual analysis of the color of the pods visually by the human eye which has limitations, which requires more energy to sort, the level of human consistency. In terms of assessing the symptoms of pests and fruit diseases, it is not guaranteed, because humans can experience fatigue, and humans also assess symptoms of pests and fruit diseases, sometimes it is subjective. This study utilizes digital image processing techniques to extract the color features of digital images of cocoa pods, the method used to extract the color features of Hue, Saturation, Value (HSV), and the classification algorithm used by K-Nearest Neighbor. The data used as many as 150 images divided into 70% training data and 30% testing data. Based on the results of trials using k values ​​of 5,7,11 and 13 in the holdout method, the best accuracy is 84.44% with a value of k = 5. And in the k-5 cross-validation test, the best accuracy is also found at k = 5 with a value accuracy of 99.33%.

Research on muscle fatigue recognition model based on improved wavelet denoising and long short-term memory

The automatic recognition technology of muscle fatigue has widespread application in the field of kinesiology and rehabilitation medicine. In this paper, we used surface electromyography (sEMG) to study the recognition of leg muscle fatigue during circuit resistance training. The purpose of this study was to solve the problem that the sEMG signals have a lot of noise interference and the recognition accuracy of the existing muscle fatigue recognition model is not high enough. First, we proposed an improved wavelet threshold function denoising algorithm to denoise the sEMG signal. Then, we build a muscle fatigue state recognition model based on long short-term memory (LSTM), and used the Holdout method to evaluate the performance of the model. Finally, the denoising effect of the improved wavelet threshold function denoising method proposed in this paper was compared with the denoising effect of the traditional wavelet threshold denoising method. We compared the performance of the proposed muscle fatigue recognition model with that of particle swarm optimization support vector machine (PSO-SVM) and convolutional neural network (CNN). The results showed that the new wavelet threshold function had better denoising performance than hard and soft threshold functions. The accuracy of LSTM network model in identifying muscle fatigue was 4.89% and 2.47% higher than that of PSO-SVM and CNN, respectively. The sEMG signal denoising method and muscle fatigue recognition model proposed in this paper have important implications for monitoring muscle fatigue during rehabilitation training and exercise.

Six skin diseases classification using deep convolutional neural network

<span>Smart imaging-based medical classification systems help the human diagnose the diseases and make better decisions about patient health. Recently, computer-aided classification of skin diseases has been a popular research area due to its importance in the early detection of skin diseases. This paper presents at its core, a system that exploits convolutional neural networks to classify color images of skin lesions. It relies on a pre-trained deep convolutional neural network to classify between six skin diseases: acne, athlete’s foot, chickenpox, eczema, skin cancer, and vitiligo. Additionally, we constructed a dataset of 3000 colored images from several online datasets and the Internet. Experimental results are encouraging, where the proposed model achieved an accuracy of 81.75%, which is higher than the state of the art researches in this field. This accuracy was calculated using the holdout method, where 90% of the images were used for training, and 10% of the images were used for out-of-sample accuracy testing.</span>

Comparison of different Classification techniques for the detection of Speech affected due to Respiratory disorders

Respiratory disorders are one of the common chronic disorders across the world. Any malfunctioning in the respiration process affects speech as speech and respiration go hand in hand. Analysis of speech parameter variation due to respiratory disorders is a prime research area in today’s scenario. In this paper, an efficient approach using a machine learning paradigm was used to detect the speech affected due to respiratory disorders. Various speech parameters were extracted (F1, F2, F3… etc.) using PRAAT software along with MFCC and LPC coefficients. Statistically, significant features were obtained to determine the predominant parameters. The extracted features were applied to different machine learning paradigms then different classification techniques were analyzed and compared to distinguish between normal and affected speech.5-fold,10-fold cross-validation, and hold-out data division protocols were used for evaluating the classification results. To evaluate the performance of the proposed methodology, Accuracy, sensitivity, specificity, and the area under receiver operating characteristics (AUC) were used. The result demonstrates that the speech parameters (F1, F2, F3…etc.) extraction method and the LPC method achieved a significantly higher classification accuracy, sensitivity, and specificity of 100% and AUC of 1, under the holdout method. While MFCC method achieved the highest classification accuracy of 83%, sensitivity of 100%, specificity of 67%, and AUC of 0.83.

Spatial modeling of soil salinity using kriging interpolation techniques: A study case in the Great Hungarian Plain

The world’s current task is to ensure food security for an ever-growing population of 7.674 billion in 2019. Soil degradation threatens sustainable agriculture in arid and semi-arid climates, where evaporation rates outweigh precipitation. Soluble salts concentrated in the subsoil under certain climatic conditions influence soil physicochemical properties, leading to soil fertility and biodiversity losses. Hence, understanding salinity behavior and its spatial variation are crucial for natural resources management to achieve and maintain sustainability. This study aims to model soil salinity spatial distribution using four kriging interpolation methods, i.e., ordinary kriging (OK), empirical Bayesian kriging (EBK), co-kriging (CK), and indicator kriging (IK). Two hundred twenty-two soil samples were collected for this purpose during a field campaign conducted in the Hungarian Soil Monitoring System framework in 2016. The performance of kriging methods was assessed and compared using two cross-validations, i.e., leave-one-out cross-validation (LOOCV) and the holdout method. The Pearson correlation analysis has been used to expose a significant moderate correlation between salt content and cation exchange capacity (CEC) with a correlation coefficient of 0.4 and a p-value of 0.003. Thus, the spatial relationship between soil salinity content (SSC) and CEC was integrated into the model to enhance predictions in areas where no measurements were accessible. The study demonstrated co-kriging efficiency by reducing the mean squared error (MSE) of ordinary kriging (OK) from 0.8 g/kg and 0.85 g/kg for LOOCV and the holdout cross-validation to 0.3 g/kg.

Exemplar Darknet19 feature generation technique for automated kidney stone detection with coronal CT images

Kidney stone is a commonly seen ailment and is usually detected by urologists using computed tomography (CT) images. It is difficult and time-consuming to detect small stones in CT images. Hence, an automated system can help clinicians to detect kidney stones accurately. In this work, a novel transfer learning-based image classification method (ExDark19) has been proposed to detect kidney stones using CT images. The iterative neighborhood component analysis (INCA) is employed to select the most informative feature vectors and these selected features vectors are fed to the k nearest neighbor (kNN) classifier to detect kidney stones with a ten-fold cross-validation (CV) strategy. The proposed ExDark19 model yielded an accuracy of 99.22% with 10-fold CV and 99.71% using the hold-out validation method. Our results demonstrate that the proposed ExDark19 detect kidney stones over 99% accuracies for two validation techniques. This developed automated system can assist the urologists to validate their manual screening of kidney stones and hence reduce the possible human error.

Real-Time Forecasting of COVID-19 Cases Using Human Mobility in Ontario, Canada

PurposeTo minimize the impact of the COVID-19 pandemic, local public health authorities are often required to make prompt and informed decisions on anticipated case-loads, resource allocation for surveillance and testing, and public health intervention appropriateness. The objective of this research was to develop a near-term forecasting model to predict COVID-19 cases using real-time human mobility information in Ontario, Canada to assist public health authorities with outbreak response.Methods & MaterialsWe utilized a deep neural network model to generate a short-term forecast of new COVID-19 cases by two weeks from May to August 2021. Variable selection was informed by a recent literature review and our ongoing research associating COVID-19 cases with human mobility, demographic and socio-economic factors. A real-time human mobility statistics consisting of a weekly summary of short and long-distance movement, demographic characteristics, weather, vaccination coverage, geo-location, and reported COVID-19 cases two weeks prior were included as predictors. We considered weeks as temporal and health regions as geographic units to account for population-level variabilities. We used a holdout method for model validation of over 300 iterations. Average mean squared error (MSE) and 95% confidence interval (CI) along with overlaying forecasted COVID-19 cases over the reported were used to evaluate the overall model fit. The model predictions were summarized as means and 95% CIs.ResultsOur best forecasting model had a mean MSE of 0.53 (95% CI: 0.49 – 0.56). Since May 2021, the overall trend of the reported COVID-19 cases in Ontario closely followed the forecasted cases, about 89% of the reported cases were within 1.5 times the interquartile range (IQR) and all were within the entire range of the distribution of the predictions. Forecasting accuracy also varied by health region characteristics, such a population size and density, remoteness, and reported COVID-19 case volume during the most recent weeks.ConclusionA near-term prediction of new COVID-19 cases with real-time population-level data could help public health authorities anticipate, plan and monitor disease burden in a population. Such predictions also allow the assessment of population-level health interventions to minimize new COVID-19 cases on a real-time basis and inform prompt decision making.

K-nearest-neighbor algorithm to predict the survival time and classification of various stages of oral cancer: a machine learning approach

Background: For years now, cancer treatments have entailed tried-and-true methods. Yet, oncologists and clinicians recommend a series of surgeries, chemotherapy, and radiation therapy. Yet, even amidst these treatments, the number of deaths due to cancer increases at an alarming rate. The prognosis of cancer patients is influenced by mutations, age, and various cancer stages. However, the association between these variables is unclear. Methods: The present work adopts a machine learning technique—k-nearest neighbor; for both regression and classification tasks, regression for predicting the survival time of oral cancer patients, and classification for classifying the patients into one of the predefined oral cancer stages. Two cross-validation approaches—hold-out and k-fold methods—have been used to examine the prediction results. Results: The experimental results show that the k-fold method performs better than the hold-out method, providing the least mean absolute error score of 0.015. Additionally, the model classifies patients into a valid group. Of the 429 records, 97 (out of 106), 99 (out of 119), 95 (out of 113), and 77 (out of 91) were classified to its correct label as stages – 1, 2, 3, and 4. The accuracy, recall, precision, and F-measure for each classification group obtained are 0.84, 0.85, 0.85, and 0.84. Conclusions: The study showed that aged patients with a higher number of mutations than young patients have a higher risk of short survival. Senior patients with a more significant number of mutations have an increased risk of getting into the last cancer stage

A novel sleep staging network based on multi-scale dual attention

Sleep is extremely important for protecting people’s mental and physical health. Once the sleep disorder occurs, people’s lives will be greatly affected. Sleep staging plays an important role in the diagnosis of sleep disorders. In general, experts classify sleep stages manually based on polysomnography (PSG), which is quite time-consuming. Meanwhile, the acquisition process of multiple signals is much complex, which can affect the subject’s sleep. Therefore, the use of single-channel electroencephalogram (EEG) for automatic sleep staging has become a popular research topic. For EEG signals, several in-siding salient waveforms used to distinguish sleep stages generally have different scales, and a single-scale convolutional neural network(CNN) cannot fully capture the salient waveforms features. To address this issue, we proposed a multiscale dual attention network (MSDAN) based on raw EEG, which utilizes a 1d CNN to automatically extract features from raw EEG. Experiments were conducted using two datasets with 20-fold cross-validation and hold-out validation method. The final average accuracy, overall accuracy, macro F1 score and Cohen’s Kappa coefficient of the model reach 96.70%, 91.74%, 0.8231 and 0.8723 on the Sleep-EDF dataset, 96.14%, 90.35%, 0.7945 and 0.8284 on the Sleep-EDFx dataset. The results show the superiority of our network over the existing methods, reaching state-of-the-arts.

Simultaneous retrieval of selected optical water quality indicators from Landsat-8, Sentinel-2, and Sentinel-3

Constructing multi-source satellite-derived water quality (WQ) products in inland and nearshore coastal waters from the past, present, and future missions is a long-standing challenge. Despite inherent differences in sensors’ spectral capability, spatial sampling, and radiometric performance, research efforts focused on formulating, implementing, and validating universal WQ algorithms continue to evolve. This research extends a recently developed machine-learning (ML) model, i.e., Mixture Density Networks (MDNs) (Pahlevan et al., 2020; Smith et al., 2021), to the inverse problem of simultaneously retrieving WQ indicators, including chlorophyll-a (Chla), Total Suspended Solids (TSS), and the absorption by Colored Dissolved Organic Matter at 440 nm (acdom(440)), across a wide array of aquatic ecosystems. We use a database of in situ measurements to train and optimize MDN models developed for the relevant spectral measurements (400–800 nm) of the Operational Land Imager (OLI), MultiSpectral Instrument (MSI), and Ocean and Land Color Instrument (OLCI) aboard the Landsat-8, Sentinel-2, and Sentinel-3 missions, respectively. Our two performance assessment approaches, namely hold-out and leave-one-out, suggest significant, albeit varying degrees of improvements with respect to second-best algorithms, depending on the sensor and WQ indicator (e.g., 68%, 75%, 117% improvements based on the hold-out method for Chla, TSS, and acdom(440), respectively from MSI-like spectra). Using these two assessment methods, we provide theoretical upper and lower bounds on model performance when evaluating similar and/or out-of-sample datasets. To evaluate multi-mission product consistency across broad spatial scales, map products are demonstrated for three near-concurrent OLI, MSI, and OLCI acquisitions. Overall, estimated TSS and acdom(440) from these three missions are consistent within the uncertainty of the model, but Chla maps from MSI and OLCI achieve greater accuracy than those from OLI. By applying two different atmospheric correction processors to OLI and MSI images, we also conduct matchup analyses to quantify the sensitivity of the MDN model and best-practice algorithms to uncertainties in reflectance products. Our model is less or equally sensitive to these uncertainties compared to other algorithms. Recognizing their uncertainties, MDN models can be applied as a global algorithm to enable harmonized retrievals of Chla, TSS, and acdom(440) in various aquatic ecosystems from multi-source satellite imagery. Local and/or regional ML models tuned with an apt data distribution (e.g., a subset of our dataset) should nevertheless be expected to outperform our global model.

Thai Named Entity Recognition Using BiLSTM-CNN-CRF Enhanced by TCC

The languages spoken in Asia share common morphological analysis errors in word segmentation which normally propagate to higher-level processing, i.e., POS tagging, syntactic parsing, word extraction, and NER, as we discuss in this research. We introduce the Thai character cluster (TCC) to reduce the errors propagated from word segmentation and POS tagging by incorporating into character representation layer of BiLSTM for NER. The initial NER model is created from the original THAI-NEST named-entity tagged corpus by applying the best performing BiLSTM-CNN-CRF model with word, part-of-speech, and character cluster embedding. We determine the errors and improve the consistency of the NE annotation through our holdout method by retraining the model with the corrected training set. After the iteration, the overall result of annotation F1-score has been improved to reach 89.22%, which improves 16.21% from the model trained on the original corpus. The result of our iterative verification is a promising method for low resource language modeling. As a result, The NE silver standard corpus is newly generated for the Thai NER task, called BKD Corpus (Bangkok Data NE tagged Corpus). The consistency of annotation is checked and revised according to the improvement of the scope of NE detection by TCC which can recover the errors in word segmentation.

A Scalable Machine Learning-based Ensemble Approach to Enhance the Prediction Accuracy for Identifying Students at-Risk

Among the educational data mining problems, the early prediction of the students' academic performance is the most important task, so that timely and requisite support may be provided to the needy students. Machine learning techniques may be used as an important tool for predicting low-performers in educational institutions. In the present paper, five single-supervised machine learning techniques have been used, including Decision Tree, Naïve Bayes, k-Nearest-Neighbor, Support Vector Machine, and Logistic Regression. To analyze the effect of an imbalanced dataset, the performance of these algorithms has been checked with and without various resampling methods such as Synthetic Minority Oversampling Technique (SMOTE), Borderline SMOTE, SVM-SMOTE, and Adaptive Synthetic (ADASYN). The Random hold-out method and GridSearchCV were used as model validation techniques and hyper-parameter tuning respectively. The results of the present study indicated that Logistic Regression is the best performing classifier with every balanced dataset generated using all of the four resampling techniques and also achieved the highest accuracy of 94.54% with SMOTE. Furthermore, to improve the prediction results and to make the model scalable, the most suitable classifier was integrated with the help of bagging, and a well-accepted accuracy of 95.45% was achieved.

Real-world prognostic factors for survival among treated patients with metastatic pancreatic ductal adenocarcinoma.

BackgroundMany real‐world studies of patients with metastatic pancreatic ductal adenocarcinoma (mPDAC) are restricted to single centers, limiting the generalizability of their insights. This study aimed to identify important population‐based predictors for survival in patients diagnosed with mPDAC in a broader setting.MethodsData between 1 January 2017 and 31 December 2019 were extracted from the Flatiron Health EHR database. Treatment‐specific predictive models were generated for patients treated with first‐line gemcitabine+nab­paclitaxel (GNP), FOLFIRINOX, gemcitabine monotherapy (gem‐mono), and second‐line liposomal irinotecan‐based regimens. The holdout method was used for cross‐validation. Age at diagnosis, sex, BMI, smoking status, and ECOG performance score were included in all models with additional demographic, clinical characteristics, and hematological function assessed for inclusion.ResultsOf the 3625 patients, 43% received GNP, 26% received FOLFIRINOX, 7% received gem‐mono, and 23% received other regimens; 40% (n = 1448) advanced to the second line. Among all first‐line patients, the following were included in the final model: prior surgery, white blood cell (WBC) counts, serum albumin (SA), liver function tests (LFTs), serum bilirubin, serum carbohydrate antigen 19–9, and ascites. Models for patients receiving specific therapies differed from the overall model, GNP (ascites removed), FOLFIRINOX (stage at initial diagnosis added), and gem‐mono (LFTs omitted). Alkaline phosphatase (ALP), SA, and WBC counts were important predictors of survival among patients treated with second‐line liposomal irinotecan. Across all regimens, the strongest predictors of survival were ECOG score, SA, and ALP.ConclusionsIn this real‐world study of patients with mPDAC, important population prognostic factors of survival were identified in a large cohort of patients receiving systemic treatment.

Clinical significance of the deep learning algorithm based on contrast-enhanced CT in the differential diagnosis of gastric gastrointestinal stromal tumors with a diameter ≤ 5 cm

Objective: Contrast-enhanced CT is an important method of preoperative diagnosis and evaluation for the malignant potential of gastric submucosal tumor (SMT). It has a high diagnostic accuracy rate in differentiating gastric gastrointestinal stromal tumor (GIST) with a diameter greater than 5 cm from gastric benign SMT. This study aimed to use deep learning algorithms to establish a diagnosis model (GISTNet) based on contrast-enhanced CT and evaluate its diagnostic value in distinguishing gastric GIST with a diameter ≤ 5 cm and other gastric SMT before surgery. Methods: A diagnostic test study was carried out. Clinicopathological data of 181 patients undergoing resection with postoperative pathological diagnosis of gastric SMT with a diameter ≤ 5 cm at Department of Gastrointestinal Surgery of Renji Hospital from September 2016 to April 2021 were retrospectively collected. After excluding 13 patients without preoperative CT or with poor CT imaging quality, a total of 168 patients were enrolled in this study, of whom, 107 were GIST while 61 were benign SMT (non-GIST), including 27 leiomyomas, 24 schwannomas, 6 heterotopic pancreas and 4 lipomas. Inclusion criteria were as follows: (1) gastric SMT was diagnosed by contrast-enhanced CT before surgery; (2) preoperative gastroscopic examination and biopsy showed no abnormal cells; (3) complete clinical and pathological data. Exclusion criteria were as follows: (1) patients received anti-tumor therapy before surgery; (2) without preoperative CT or with poor CT imaging quality due to any reason; (3) except GIST, other gastric malignant tumors were pathologically diagnosed after surgery. Based on the hold-out method, 148 patients were randomly selected as the training set and 20 patients as the test set of the GISTNet diagnosis model. After the GISTNet model was established, 5 indicators were used for evaluation in the test set, including sensitivity, specificity, positive predictive value, negative predictive value and the area under the receiver operating curve (AUC). Then GISTNet diagnosis model was compared with the GIST-risk scoring model based on traditional CT features. Besides, in order to compare the accuracy of the GISTNet diagnosis model and the imaging doctors in the diagnosis of gastric SMT imaging, 3 radiologists with 3, 9 and 19 years of work experience, respectively, blinded to clinical and pathological information, tested and judged the samples. The accuracy rate between the three doctors and the GISTNet model was compared. Results: The GISTNet model yielded an AUC of 0.900 (95% CI: 0.827-0.973) in the test set. When the threshold value was 0.345, the sensitivity specificity, positive and negative predictive values of the GISTNet diagnosis model was 100%, 67%, 75% and 100%, respectively. The accuracy rate of the GISTNet diagnosis model was better than that of the GIST-risk model and the manual readings from two radiologists with 3 years and 9 years of work experience (83% vs. 75%, 60%, 65%), and was close to the manual reading of the radiologist with 19 years of work experience (83% vs. 80%). Conclusion: The deep learning algorithm based on contrast-enhanced CT has favorable and reliable diagnostic accuracy in distinguishing gastric GIST with a diameter ≤ 5 cm and other gastric SMT before operation.

A personalized blood glucose level prediction model with a fine-tuning strategy: A proof-of-concept study

BackgroundThe accurate prediction of blood glucose (BG) level is still a challenge for diabetes management. This is due to various factors such as diet, personal physiological characteristics, stress, and activities influence changes in BG level. To develop an accurate BG level predictive model, we propose a personalized model based on a convolutional neural network (CNN) with a fine-tuning strategy. MethodsWe utilized continuous glucose monitoring (CGM) datasets from 1052 professional CGM sessions and split them into three groups according to type 1, type 2, and gestational diabetes mellitus (T1DM, T2DM, and GDM, respectively). During the preprocessing, only CGM data points were utilized, and future BG levels of four different prediction horizons (PHs, 15, 30, 45, and 60 min) were used as output. In training, we trained a general CNN and a multi-output random forest regressor using a hold-out method for each group. Next, we developed two personalized models: (1) by fine-tuning the general CNN on partial sample points of each CGM dataset, and (2) by learning a CNN from scratch on the points. ResultsFor all groups, the fine-tuned CNN showed the lowest average root mean squared error, average mean absolute percentage error, highest average time gain (PH = 15 and 60 min in T1DM) and highest percentage in region A of Clarke error grid analysis at all PHs. In the performance comparison between the fine-tuned CNN and other models, we found that the fine-tuned CNN improved the performance of the general CNN in most cases and outperformed the scratch CNN at all PHs in all groups, making the fine-tuning strategy was useful for accurate BG level prediction. We analyzed all cases of four predictive patterns in each group, and found that the input BG level trend and the BG level at the time of prediction were related to the future BG level trend. ConclusionsWe demonstrated the efficacy of the fine-tuning method in a large number of CGM datasets and analyzed the four predictive patterns. Therefore, we believe that the proposed method will significantly contribute to the development of an accurate personalized model and the analysis for its predictions.

SISTEM PREDIKSI RASA BUAH JERUK MENGGUNAKAN METODE k-NEAREST NEIGHBOR

Every buyer always wants to get a good product, including in terms of buyers of citrus fruits, but many buyers do not know how to choose sweet oranges if they have not been tested first, so buyers usually experience disappointment when buying citrus fruits. The purpose of this study is to assist buyers in choosing citrus fruits, the k-NN method in identifying the taste of citrus fruits. The results of testing with the k-fold cross-validation method obtained a value of 96.55%, while testing with the holdout method obtained a value of 91.67%. The benefit of this research is that by using the application that was built, buyers can predict the taste of citrus fruits without having to taste them first.

A novel structural adaptive discrete grey prediction model and its application in forecasting renewable energy generation

The rapidly growing renewable energy generation instigates stochastic volatility of electricity supply that may compromise the power grid's stability and increase the grid imbalance cost. Therefore, accurate mid-to-long term renewable energy generation forecasting is of great significance for integrating renewable energy systems with smart grid and energy strategic planning. For this purpose, a new structural adaptive discrete grey prediction model is proposed. Overall, the proposed model possesses three main contributions. Firstly, the introduction of nonlinear term and periodic term strengthens the ability of the traditional DGM (1,1) model to capture the nonlinear and linear development trend of time series and improves the adaptability of the grey prediction model to arbitrary periodic time series. Secondly, the emerging coefficients are determined by the particle swarm optimization algorithm and hold-out cross-validation method, and the adaptive selection of the model structure is realized. From the perspective of expert system, it reduces the need for modeling knowledge. Thirdly, the consistency of stretching, unbiasedness, and compatibility with other grey models are discussed, which further verified the feasibility and practicability of the proposed model. Besides, the performance of the proposed model is compared with those of a series of grey prediction models and non-grey prediction methods to verify the feasibility and superiority of this new approach by three real cases. The results indicate that the proposed model benefits from its adaptive structure and produces reliable predictions.

Deep Feature Extraction for Detection of Tomato Plant Diseases and Pests based on Leaf Images

Plant diseases and pests cause yield and quality losses. It has great importance to detect plant diseases and pests quickly and with high accuracy in terms of preventing yield and quality losses. Plant disease and pest detection performed by plant protection experts through visual observation is a labor-intensive process with a high error rate. Developing effective, fast and highly successful computer-aided disease detection systems has become a necessity in terms of precision agriculture applications. In this study, well-known pre-trained convolutional neural network (CNN) models AlexNet, GoogLeNet and ResNet-50 are used as feature extractors. In addition, a deep learning model that concatenate deep features extracted from 3 CNN models has been proposed. The deep features were used to train the support vector machine classifier. The proposed model was used to classify leaf images of tomato plant diseases and pests, which is a subset of open-access PlantVillage dataset consisting of a total of 18835 images belonging to 10 classes including a healthy one. Accuracy, precision, sensitivity and f-score performance metrics were used with the hold-out validation method in determining model performances. Experimental results show that the detection of tomato plant diseases and pests is possible using concatenated deep features with an overall accuracy rate of 96.99%.

The Classification of Taekwondo Kicks Via Machine Learning: A feature selection investigation

Martial art strike classification by machine learning has drawn more attention over the past decade. The unique signal from each technique makes it harder to be recognized. Thus, this paper proposed an SVM, Random Forest, k-NN, and Naïve Bayes classification method applied to the time-domain signal to classify the three type of taekwondo technique. Data collected from the total of five participant and statistical features such as mean, median, minimum, maximum, standard deviation, variance, skewness, kurtosis, and standard error mean were extracted from the signal. After that, the data will be trained using selected rank features and hold-out method with k-fold cross-validation applied to the training and testing data. Therefore, with ANOVA test as features selection and 60:40 ratio of a hold-out method, Random Forest classifier score the highest accuracy of 86.7%..

Prediction of the breakthrough cancer pain using deep learning model.

e24085 Background: Breakthrough cancer pain (BTcP), a transitory flare of pain that occurs on a background of relatively well-controlled baseline pain, is a challenging clinical problem in managing cancer pain. We hypothesized that the BTcP could be predictable according to the patients’ previous observed patterns. In this study, we report on the development of a deep learning model that predicts hourly individual-level breakthrough pain for patients with cancer. Methods: We defined the BTcP as the pain with numerical rating scale (NRS) score 4 or above and developed models predicting the onset time of BTcP with the temporal resolution of 1 hour. The datasets which have more than 20 records of NRS score during hospitalization were included in our study. All the pain records were obtained from patients hospitalized on the wards of hematology-oncology in Samsung Medical Center between July 2016 to February 2020. The model used the time windows of 3 days to predict NRS scores over the next 24 hours. To capture irregular pain patterns, we created the sequence of average pain patterns over 24 hours from the previous 3 days and used it for normalization. We trained a Bi-directional long-short term memory (LSTM) based deep learning model. The model was validated using the holdout method with 20% of the datasets. Its performance was assessed with the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AURPC). Results: We included pain log data containing 2,905 admissions from 2,176 patients with solid cancer and 1,755 admissions from 1,082 patients with hematologic cancer in the analysis. The median age was 57 (interquartile range (IQR), 47-64), the most frequent type of cancer was lung cancer (18.0%), and most patients had stage 4 (60.7%). Among the 103,948 hours from patients in whole datasets, 1,091 (4.7%) hours were labeled as the period of BTcP. The patients have the records of NRS score with a median of 3 (IQR, 2.0-4.5) and BTcP with a median of 1.1 (IQR, 0.5-2.0) per day. We allocated approximately 20% of patients (653 patients with 932 admissions) to the holdout test dataset. Our model showed the AUROC 0.719 and AUPRC 0.680 for predicting the BTcP in the test dataset. Conclusions: Our study showed that cancer pain could be predictive by using a deep learning model. Though our exploratory study has a limitation of generalizability, future warranted subgroup analysis and verification research could make our model more applicable in a real-world setting.