Holdout Method Research Articles

Prognostic prediction of cardiovascular adverse events in patients after percutaneous coronary intervention using machine learning

Abstract Introduction While prevention of major cardiovascular adverse events (MACE) is important in the management of patients after percutaneous coronary intervention (PCI), risk stratification of MACE remains a significant challenge with increasingly diverse and complex patient backgrounds. In response to that, this retrospective observational study aimed to explore whether machine learning (ML) methods can predict future MACE after PCI using a variety of features extracted from electronic medical records (EMRs). Methods In 3,016 patients with coronary artery disease who performed PCI, 3 machine learning (ML) algorithms (gentle boost, random forest, and neural network) and statistical model were retrospectively applied to predict the occurrence of MACE within two years after PCI (564 patients, 18.7%). MACE was defined as composite endpoint of cardiac death, myocardial infarction (MI), any revascularization, and congestive heart failure (CHF). Basic patient information (e.g., age, sex, coronary risk factors) and laboratory test results were used for the analysis. Feature importance was calculated to identify key factors for MACE. Hold-out method with an 8:2 split between training and test was used to calculate sensitivity, specificity, positive predictive value, negative predictive value, and F-value for each model. Results As shown in Figure 1, ML models showed excellent diagnostic accuracies in the prediction of MACE after PCI, demonstrating higher F-values compared with statistical model (logistic regression analysis). Similar results were seen in the subgroup analyses for estimating cardiac death, MI, revascularization, or CHF. Although the weighting of feature importance for MACE recurrence varied among ML models, infarct size (peak CK levels), increased inflammatory status, and traditional coronary risk factors were identified as important features for predicting MACE within two years after PCI (Figure 2). Conclusions It is suggested that ML using EMRs is promising for inferring future MACE within two years after PCI. Further studies are warranted to address whether ML-based risk stratification could lead to improved clinical decision-making after PCI.

Daily peak demand forecasting using Pelican Algorithm optimised Support Vector Machine (POA-SVM)

The knowledge of daily peak load consumption is crucial for energy planning, energy management, and resource allocation, as it is an essential element of supply-side management. This knowledge is obtained from accurate predictions of which traditional methods are falling short due to constantly changing demand. Hence, there is a need for Machine Learning (ML) models that are more efficient in hybridised or enhanced forms. This paper presents the use of a novel Pelican Algorithm optimised Support Vector Machine (POA-SVM), a hybridised ML Algorithm, to predict peak load and its corresponding peak hour for proper planning. It evaluates the performance of four Support Vector Machine (SVM) models: standard SVM, SVM optimised with Bayesian Optimisation (SVMB), SVM optimised with Particle Swarm Optimization (PSO-SVM) and POA-SVM on both raw and normalised datasets. Its analysis includes a comparison of performance metrics such as Root Mean Squared Error (RMSE), Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared (R²) to determine the models' accuracy and goodness of fit. Furthermore, we experiment with K-fold cross-validation and the hold-out validation method, finding that K-fold cross-validation yields better performance metrics for all models. Notably, the POA-SVM model shows superior performance across all metrics, particularly on the raw dataset, making it a robust choice for forecasting peak demand and its corresponding hour of the day.

Fault Diagnosis of Wind Turbine Blades with Continuous Wavelet Transform based Deep Learning Model Using Vibration Signal

Wind Turbines are the most crucial devices in wind energy conversion systems to increase energy generation efficiency from wind sources. Blade failure in wind turbines is due to the induced vibrations of change in ecological variables. It is necessary to periodically inspect the state of the wind turbine blades to enhance safety by decreasing downtime, lessening the likelihood of unplanned failures requiring extensive repair, and progressively increasing power generation with reduced logistical costs. The primary goal of the proposed research is to adopt a hybrid strategy that combines convolutional neural network models and continuous wavelet transform. The bump wavelet-based continuous wavelet transform with a convolutional neural network model is employed to classify the faulty wind turbine blades based on the extracted vibration signals of turbine blades. This approach distinguishes between different states of blade faults affecting wind turbine blades during operational phases. The research considers blade faults in the horizontal axis wind turbine, including blade bending, erosion, and the connection looseness between the hub and blade. The cross-validation and hold-out methods are used to validate the classification accuracy and are compared with other existing popular methods. The hold-out method has a better classification accuracy of 97.916%.

Estimation of Blood Calcium and Potassium Values from ECG Records

Abstract The identification of diseases caused by changes in ion concentration is quite difficult and yet plays a decisive role in the success of clinical care, diagnosis and treatment. The clinically proven approach to diagnosing electrolyte concentration imbalance is blood tests. There is a need to provide a non-invasive diagnostic method that is not of a temporary nature. Bio-signals such as the electrocardiogram (ECG) can be used to meet this demand and become diagnostic tools that facilitate home monitoring of electrolyte concentration on a permanent basis. This study investigates the feasibility and efficiency of methods based on machine learning (ML) and ECG recordings in monitoring critical levels of existing potassium and calcium concentration. Morphological, frequency and frequency-time domain features were extracted to automatically estimate calcium and potassium levels. Furthermore, the potential of estimates based on modeling approaches will be demonstrated to gain insights into relevant clinical findings and improve the performance of monitoring approaches. Using the hold-out validation method, the best results in terms of mean square error (MSE) and R for estimating the calcium value are 0.7157 and 0.57347, using fuzzy inference systems (FIS). Here, R represents the proportion of the variance in the calcium value that is explained by the model.

Optimizing Fall Risk Diagnosis in Older Adults Using a Bayesian Classifier and Simulated Annealing.

The aim of this study was to improve the diagnostic ability of fall risk classifiers using a Bayesian approach and the Simulated Annealing (SA) algorithm. A total of 47 features from 181 records (40 Center of Pressure (CoP) indices and 7 patient descriptive variables) were analyzed. The wrapper method of feature selection using the SA algorithm was applied to optimize the cost function based on the difference of the mean minus the standard deviation of the Area Under the Curve (AUC) of the fall risk classifiers across multiple dimensions. A stratified 60-20-20% hold-out method was used for train, test, and validation sets, respectively. The results showed that although the highest performance was observed with 31 features (0.815 ± 0.110), lower variability and higher explainability were achieved with only 15 features (0.780 ± 0.055). These findings suggest that the SA algorithm is a valuable tool for feature selection for acceptable fall risk diagnosis. This method offers an alternative or complementary resource in situations where clinical tools are difficult to apply.

Harnessing Artificial Intelligence to Predict Ovarian Stimulation Outcomes in In Vitro Fertilization: Scoping Review.

In the realm of in vitro fertilization (IVF), artificial intelligence (AI) models serve as invaluable tools for clinicians, offering predictive insights into ovarian stimulation outcomes. Predicting and understanding a patient's response to ovarian stimulation can help in personalizing doses of drugs, preventing adverse outcomes (eg, hyperstimulation), and improving the likelihood of successful fertilization and pregnancy. Given the pivotal role of accurate predictions in IVF procedures, it becomes important to investigate the landscape of AI models that are being used to predict the outcomes of ovarian stimulation. The objective of this review is to comprehensively examine the literature to explore the characteristics of AI models used for predicting ovarian stimulation outcomes in the context of IVF. A total of 6 electronic databases were searched for peer-reviewed literature published before August 2023, using the concepts of IVF and AI, along with their related terms. Records were independently screened by 2 reviewers against the eligibility criteria. The extracted data were then consolidated and presented through narrative synthesis. Upon reviewing 1348 articles, 30 met the predetermined inclusion criteria. The literature primarily focused on the number of oocytes retrieved as the main predicted outcome. Microscopy images stood out as the primary ground truth reference. The reviewed studies also highlighted that the most frequently adopted stimulation protocol was the gonadotropin-releasing hormone (GnRH) antagonist. In terms of using trigger medication, human chorionic gonadotropin (hCG) was the most commonly selected option. Among the machine learning techniques, the favored choice was the support vector machine. As for the validation of AI algorithms, the hold-out cross-validation method was the most prevalent. The area under the curve was highlighted as the primary evaluation metric. The literature exhibited a wide variation in the number of features used for AI algorithm development, ranging from 2 to 28,054 features. Data were mostly sourced from patient demographics, followed by laboratory data, specifically hormonal levels. Notably, the vast majority of studies were restricted to a single infertility clinic and exclusively relied on nonpublic data sets. These insights highlight an urgent need to diversify data sources and explore varied AI techniques for improved prediction accuracy and generalizability of AI models for the prediction of ovarian stimulation outcomes. Future research should prioritize multiclinic collaborations and consider leveraging public data sets, aiming for more precise AI-driven predictions that ultimately boost patient care and IVF success rates.

Ultrasound-based radiomics machine learning models for diagnosing cervical lymph node metastasis in patients with non-small cell lung cancer: a multicentre study

BackgroundCervical lymph node metastasis (LNM) is an important prognostic factor for patients with non-small cell lung cancer (NSCLC). We aimed to develop and validate machine learning models that use ultrasound radiomic and descriptive semantic features to diagnose cervical LNM in patients with NSCLC.MethodsThis study included NSCLC patients who underwent neck ultrasound examination followed by cervical lymph node (LN) biopsy between January 2019 and January 2022 from three institutes. Radiomic features were extracted from the ultrasound images at the maximum cross-sectional areas of cervical LNs. Logistic regression (LR) and random forest (RF) models were developed. Model performance was assessed by the area under the curve (AUC) and accuracy, validated internally and externally by fivefold cross-validation and hold-out method, respectively.ResultsIn total, 313 patients with a median age of 64 years were included, and 276 (88.18%) had cervical LNM. Three descriptive semantic features, including long diameter, shape, and corticomedullary boundary, were selected by multivariate analysis. Out of the 474 identified radiomic features, 9 were determined to fit the LR model, while 15 fit the RF model. The average AUCs of the semantic and radiomics models were 0.876 (range: 0.781–0.961) and 0.883 (range: 0.798–0.966), respectively. However, the average AUC was higher for the semantic-radiomics combined LR model (0.901; range: 0.862–0.927). When the RF algorithm was applied, the average AUCs of the radiomics and semantic-radiomics combined models were improved to 0.908 (range: 0.837–0.966) and 0.922 (range: 0.872–0.982), respectively. The models tested by the hold-out method had similar results, with the semantic-radiomics combined RF model achieving the highest AUC value of 0.901 (95% CI, 0.886–0.968).ConclusionsThe ultrasound radiomic models showed potential for accurately diagnosing cervical LNM in patients with NSCLC when integrated with descriptive semantic features. The RF model outperformed the conventional LR model in diagnosing cervical LNM in NSCLC patients.

Applying Recurrent Neural Networks and Blocked Cross-Validation to Model Conventional Drinking Water Treatment Processes

The jar test is the current standard method for predicting the performance of a conventional drinking water treatment (DWT) process and optimizing the coagulant dose. This test is time-consuming and requires human intervention, meaning it is infeasible for making continuous process predictions. As a potential alternative, we developed a machine learning (ML) model from historical DWT plant data that can operate continuously using real-time sensor data without human intervention for predicting clarified water turbidity 15 min in advance. We evaluated three types of models: multilayer perceptron (MLP), the long short-term memory (LSTM) recurrent neural network (RNN), and the gated recurrent unit (GRU) RNN. We also employed two training methodologies: the commonly used holdout method and the theoretically correct blocked cross-validation (BCV) method. We found that the RNN with GRU was the best model type overall and achieved a mean absolute error on an independent production set of as low as 0.044 NTU. We further found that models trained using BCV typically achieve errors equal to or lower than their counterparts trained using holdout. These results suggest that RNNs trained using BCV are superior for the development of ML models for DWT processes compared to those reported in earlier literature.

Combining Sentinel 1, Sentinel 2 and MODIS data for major winter crop type classification over the Murray Darling Basin in Australia

Crop type classification is an essential task in agriculture that has been studied widely since the emergence of remote sensing technologies. Accurate crop mapping can help assist decision-making related to storage, marketing, and other production tasks. This study has designed robust crop classification models to classify two major crop types (cereals and canola) in the Murray Darling Basin (MDB) in Australia. These models combined Sentinel 1 and 2 and Moderate Resolution Imaging Spectroradiometer (MODIS) data. Three methods were applied to test classification quality: the holdout method, leave-one-season-out cross-validation (LOSOCV), and leave-one-cluster-out-cross-validation (LOCOCV). The holdout method evaluated the model's performance on data representing the entire population. The LOSOCV method was used to test the model's ability to extrapolate over time (unseen data from new seasons), and the LOCOCV method was used to test the ability to extrapolate over space (data from a new site). Crop type labelled data (n = 193 for cereals and n = 113 for canola) were extracted from high-resolution yield maps derived from grain yield monitors mounted on harvesters, and this was used to train and validate the classification models. The results showed that the holdout validation approach achieved the highest accuracy (overall and Kappa scores >0.98), but this might be caused by spatial autocorrelation of the sampling strategy implemented, which leads to an overoptimistic scenario. The overall accuracies and Kappa scores achieved from the LOSOCV method varied depending upon the difference in crop phenology from season to season, with overall accuracy ranging from 0.70 to 0.92 and Kappa scores from 0.61 to 0.90. The overall accuracies and Kappa scores for the LOCOCV ranged from 0.87 to 0.99 and 0.84 to 0.98, respectively. The gross primary productivity (GPP) and the chlorophyll red edge index (CIr) were the most important features across the three models. The outputs of this research can potentially complement models that require crop type information, such as carbon cycling and crop yield models.

A novel time-varying grey Fourier model for variable amplitude seasonal fluctuation sequences

PurposeA time-varying grey Fourier model (TVGFM(1,1,N)) is proposed for the simulation of variable amplitude seasonal fluctuation time series, as the performance of traditional grey models can't catch the time-varying trend well.Design/methodology/approachThe proposed model couples Fourier series and linear time-varying terms as the grey action, to describe the characteristics of variable amplitude and seasonality. The truncated Fourier order N is preselected from the alternative order set by Nyquist-Shannon sampling theorem and the principle of simplicity, then the optimal Fourier order is determined by hold-out method to improve the robustness of the proposed model. Initial value correction and the multiple transformation are also studied to improve the precision.FindingsThe new model has a broader applicability range as a result of the new grey action, attaining higher fitting and forecasting accuracy. The numerical experiment of a generated monthly time series indicates the proposed model can accurately fit the variable amplitude seasonal sequence, in which the mean absolute percentage error (MAPE) is only 0.01%, and the complex simulations based on Monte-Carlo method testify the validity of the proposed model. The results of monthly electricity consumption in China's primary industry, demonstrate the proposed model catches the time-varying trend and has good performances, where MAPEF and MAPET are below 5%. Moreover, the proposed TVGFM(1,1,N) model is superior to the benchmark models, grey polynomial model (GMP(1,1,N)), grey Fourier model (GFM(1,1,N)), seasonal grey model (SGM(1,1)), seasonal ARIMA model seasonal autoregressive integrated moving average model (SARIMA) and support vector regression (SVR).Originality/valueThe parameter estimates and forecasting of the new proposed TVGFM are studied, and the good fitting and forecasting accuracy of time-varying amplitude seasonal fluctuation series are testified by numerical simulations and a case study.

A discrete time-varying grey Fourier model with fractional order terms for electricity consumption forecast

Accurate electricity consumption forecast plays an important role in managing and operating a country’s power system and is a challenging task due to the data exhibiting the mixture characteristics of trend, periodicity and randomness. To accurately describe the dynamic changes of seasonal fluctuations of electricity consumption, a novel discrete time-varying grey Fourier model with fractional order terms is proposed by coupling Fourier functions with fractional order time-varying terms, as the grey action, to fit the amplitude variations hidden in seasonal pattern and improve the fitting and forecasting accuracy. Moreover, the mechanism and properties of the proposed model are discussed. Then the optimal fractional order is determined by the Dingo Optimization Algorithm (DOA) and the optimal Fourier order is selected by the hold-out method to enhance the robustness of the model. Subsequently, the results of designed experiments based on Monte-Carlo method testify the validity of the Fourier order selection method and the proposed model. Finally, this model is applied to predict the monthly residential electricity consumption of China from 2015 to 2022. The results indicate that the proposed model captures the dynamic amplitude variations of the time series and has better prediction performances than other benchmark grey prediction models and non-grey prediction models.

DEEPFAKE FACE RECOGNITION THROUGH MODIFIED AND IMPORVED DEEP TRANSFER LEARNING

The manipulation of photo, audio, and video content has been a topic of interest for many years, as people uses fake faces to indulge in various immoral act like pornography, fraud, and defamation. In the early days, classification of real and fake faces was done using traditional methods such as editing frames by frame or using chroma keying, this traditional approach is time consuming and lacks enough editors that have the technical skills to do the frame-by-frame edition or use the chroma keying. With technological advancement, new techniques have been developed that allow for much more sophisticated and realistic manipulations, one such technique is deepfakes. Deepfakes are created using deep learning algorithms to swap or replace faces in videos or images. This can be done with a high degree of realism, making it difficult to distinguish between real and fake content. This research aims to develop a deep fake detection system using deep transfer learning (modified VGG19 and ResNet50 models), these two models were chosen over other CNN architectures due to their proven better performance, faster recognition time and lesser memory usage. The research modified the original VGG19 and ResNet architectures by replacing the last five layer with a customized dense layers that will help with faster and accurate recognition of faces. A balanced dataset comprising 70,000 real faces from the Flickr dataset and 70,000 fake faces generated by StyleGAN was utilized. This research employed hold-out evaluation method. VGG19 gave an accuracy, f1score of 91.59% and 91.47% respectively while RestNet50 gave an average accuracy and F1score of 96.61% and 96.59% respectively on the testing dataset. This shows that ResNet 50 gave the best performance both on the training, validation and testing dataset. The developed system was also compared with other state-of-art methods and they were all outperformed.

Explanation of the influence of geomorphometric variables on the landform classification based on selected areas in Poland

In recent years, automatic image classification methods have significantly progressed, notably black box algorithms such as machine learning and deep learning. Unfortunately, such efforts only focused on improving performance, rather than attempting to explain and interpret how classification models actually operate. This article compares three state-of-the-art algorithms incorporating random forests, gradient boosting and convolutional neural networks for geomorphological mapping. It also attempts to explain how the most effective classifier makes decisions by evaluating which of the geomorphometric variables are most important for automatic mapping and how they affect the classification results using one of the explainable artificial intelligence techniques, namely accumulated local effects (ALE). This method allows us to understand the relationship between predictors and the model’s outcome. For these purposes, eight sheets of the digital geomorphological map of Poland on the scale of 1:100,000 were used as the reference material. The classification results were validated using the holdout method and cross-validation for individual sheets representing different morphogenetic zones. The terrain elevation entropy, absolute elevation, aggregated median elevation and standard deviation of elevation had the greatest impact on the classification results among the 15 geomorphometric variables considered. The ALE analysis was conducted for the XGBoost classifier, which achieved the highest accuracy of 92.8%, ahead of Random Forests at 84% and LightGBM at 73.7% and U-Net at 59.8%. We conclude that automatic classification can support geomorphological mapping only if the geomorphological characteristics in the predicted area are similar to those in the training dataset. The ALE plots allow us to analyze the relationship between geomorphometric variables and landform membership, which helps clarify their role in the classification process.

Toward Generalizable Machine Learning Models in Speech, Language, and Hearing Sciences: Estimating Sample Size and Reducing Overfitting.

Many studies using machine learning (ML) in speech, language, and hearing sciences rely upon cross-validations with single data splitting. This study's first purpose is to provide quantitative evidence that would incentivize researchers to instead use the more robust data splitting method of nested k-fold cross-validation. The second purpose is to present methods and MATLAB code to perform power analysis for ML-based analysis during the design of a study. First, the significant impact of different cross-validations on ML outcomes was demonstrated using real-world clinical data. Then, Monte Carlo simulations were used to quantify the interactions among the employed cross-validation method, the discriminative power of features, the dimensionality of the feature space, the dimensionality of the model, and the sample size. Four different cross-validation methods (single holdout, 10-fold, train-validation-test, and nested 10-fold) were compared based on the statistical power and confidence of the resulting ML models. Distributions of the null and alternative hypotheses were used to determine the minimum required sample size for obtaining a statistically significant outcome (5% significance) with 80% power. Statistical confidence of the model was defined as the probability of correct features being selected for inclusion in the final model. ML models generated based on the single holdout method had very low statistical power and confidence, leading to overestimation of classification accuracy. Conversely, the nested 10-fold cross-validation method resulted in the highest statistical confidence and power while also providing an unbiased estimate of accuracy. The required sample size using the single holdout method could be 50% higher than what would be needed if nested k-fold cross-validation were used. Statistical confidence in the model based on nested k-fold cross-validation was as much as four times higher than the confidence obtained with the single holdout-based model. A computational model, MATLAB code, and lookup tables are provided to assist researchers with estimating the minimum sample size needed during study design. The adoption of nested k-fold cross-validation is critical for unbiased and robust ML studies in the speech, language, and hearing sciences. https://doi.org/10.23641/asha.25237045.

LIBS combined with SG-SPXY spectral data pre-processing for cement raw meal composition analysis.

Rapid testing of cement raw meal plays a crucial role in the cement production process, so there is an urgent need for a fast and accurate testing method. In this paper, a method based on the Savitzky-Golay (SG) smoothing and sample set partitioning based on joint x-y distance (SPXY) spectral data pre-processing is proposed to improve the accuracy of the laser-induced breakdown spectroscopy (LIBS) technique for quantitative analysis of cement raw meal components. Firstly, the spectral data is denoised by SG smoothing, which effectively reduces the noise and baseline variations in the spectra. Then, the denoised data is divided into sample sets by combining the SPXY sample division method, which improves the efficiency of data analysis. Finally, the delineated data set is modeled for quantitative analysis by a back-propagation (BP) neural network. Compared to the modeling effect of the four oxide contents of CaO, S i O 2, A l 2 O 3, and F e 2 O 3 in the Hold-Out method, the correlation coefficient (R) was improved by 26%, 10%, 17%, and 4%, respectively. The root mean square error (RMSE) was reduced by 47%, 33%, 43%, and 21%, respectively. The mean absolute percentage error (MAPE) was reduced by 63%, 60%, 36%, and 51%, respectively. The results show that there is a significant improvement in the model effect, which can effectively improve the accuracy of quantitative analysis of cement raw meal composition by LIBS. This is of great significance for the real-time detection of cement raw meal composition analysis.

Efficient Use of Data for Prediction and Validation

Prediction model building is one of the most important tasks in analysis of high-dimensional data. A fitted prediction model should be validated for future use. So, when conducting such an analysis, we have to use the whole data for both training and validation. When using a hold-out method, the fitted prediction model will be more efficient if the training set is bigger, but the validation power will be lower with a smaller validation set. In order to balance the efficiency of fitted prediction model and its validation, 50-50 allocation of the whole data set is popularly used as a hold-out method. In prediction and validation procedure, we have to use the information embedded in the whole data set as efficiently as possible. As a such effort, cross-validation methods (CV) have been very popular these days. In a CV method, a large portion of the data set is used to train models and the remaining small portion of the data is used for validation, and this procedure is repeated until the whole data set is used for validation. In a CV method, each data point is used for both training and validation, so that as the portion of training set is increased, the efficiency of training will be increased, while the validation power will be decreased due to the increased over-fitting, i.e. more frequent use of each data point for training. As another effort of efficient use of the whole data, we propose to use the whole data set for both training and validation, called 1-fold CV method. By using the whole data to fit a prediction model, training efficiency will be highest, but, by reusing the whole data set for validation, its validation power is expected to be very low. The validation power of CV methods will be estimated by permutation methods. Through extensive simulation and real data studies, we conclude that the newly proposed 1-fold CV method uses the available data set very efficiently

Deep machine learning identified fish flesh using multispectral imaging

Food fraud is widespread in the aquatic food market, hence fast and non-destructive methods of identification of fish flesh are needed. In this study, multispectral imaging (MSI) was used to screen flesh slices from 20 edible fish species commonly found in the sea around Yantai, China, by combining identification based on the mitochondrial COI gene. We found that nCDA images transformed from MSI data showed significant differences in flesh splices of the 20 fish species. We then employed eight models to compare their prediction performances based on the hold-out method with 70% training and 30% test sets. Convolutional neural network (CNN), quadratic discriminant analysis (QDA), support vector machine (SVM), and linear discriminant analysis (LDA) models perform well on cross-validation and test data. CNN and QDA achieved more than 99% accuracy on the test set. By extracting the CNN features for optimization, a very high degree of separation was obtained for all species. Furthermore, based on the Gini index in RF, 11 bands were selected as key classification features for CNN, and an accuracy of 98% was achieved. Our study developed a successful pipeline for employing machine learning models (especially CNN) on MSI identification of fish flesh, and provided a convenient and non-destructive method to determine the marketing of fish flesh in the future.

Detection of Papilledema Severity from Color Fundus Images using Transfer Learning Approaches

Papilledema is edema in the area where the optic nerve meets the eye as a result of increased pressure inside the head. This disease can result in very serious problems, such as abnormal optical changes, decreased visual acuity, and even permanent blindness if left untreated. In this study, an image processing based solution was presented for the detection of papilledema severity from color fundus images using transfer learning approaches. The image dataset includes 295 papilledema images, 295 pseudopapilledema images, and 779 control images. Histogram equalization and the 3D box filter were used for image preprocessing. The images were enhanced with the histogram equalization method and denoised with the 3D box filter method. Then, the performances of EfficentNet-B0, GoogLeNet, MobileNetV2, NASNetMobile, and ResNet-101 transfer learning approaches were compared. The hold-out method was used to calculate the performance of transfer learning. In the experiments, the MobileNetV2 approach had the highest performance with 0.96 overall accuracy and 0.94 Cohen's Kappa. The results of the experiments proved that the combination of the histogram equalization, the 3D box filter, and the MobileNetV2 transfer learning approach can be used for automatic detection of papilledema severity. Compared to other similar studies that are known in the literature, the overall accuracy was higher.

Categorization of Indian residential consumers electrical energy consumption pattern using clustering and classification techniques

Data Mining techniques are widely applied in power sector to analyze the electricity consumption pattern(ECP) of consumers for implementing effective demand response management. In this study, the ECP of LV consumers of Karur city is used for evaluation. The dataset consists of five years ECP. In this study, K-means clustering algorithm is used to categorize the data into binary and three classes. The results of the K-means clustering technique is used for developing eight classifier models. The hyperparameters of the classifier models are tuned using grid search optimization technique. The classification process applies hold-out partitioning and stratified 10-fold cross-validation(CV) techniques to evaluate the classifier’s efficiency in predicting a consumer’s ECP. The experimental result shows that the K-means clustering has minimal overlapping of 0.48% and 1.98% for binary and three-class clustering processes. Hence, the K-means algorithm results are used to build this study’s classification model. In the classification process, the experimental results of LR and SVM algorithms have better accuracy in binary and three class categories respectively, for the hold-out partitioning method. Meanwhile, the SVM also has better generalization ability and is validated using stratified 10-fold CV, achieving an accuracy of 99.09% and 98.19% for binary and three classes respectively, in the classification of consumer data.

The globalized fish Industry: Employing DNA-barcoding and NIRS technology to combat counterfeiting and safeguard traditional agro-food products

Traditional food plays a crucial role in showcasing the culture and heritage of its producers. The identification of the Traditional agro-Food Product (TFP) “Seppia bianca di Chioggia”, which is derived from Italian traditional cuttlefish, was performed through genetic analysis utilizing mitochondrial cytochrome c oxidase I subunit gene (COI) sequences and near-infrared spectroscopy (NIRS). The set samples were obtained from the shores of Chioggia (Adriatic Sea), and a comparison was made with cuttlefishes collected from Cesenatico (Adriatic Sea) and the coastal regions of France (Atlantic Ocean). The primary objective was to differentiate the TFP from samples sourced from nearby production areas and distant geographic locations that are commonly sold in Italy. A total of 368 cuttlefishes were analysed for NIRS data, whereas a subset of 75 samples were randomly selected and analysed for COI sequences to assess cuttlefish origin and TFP certification. Genetic and NIRS models were then tested on the same set of samples using a machine learning approach through the hold-out validation method. Additionally, an external validation was performed using a new set of cuttlefishes (labelled fished in shore of Chioggia or in Atlantic Ocean) collected from the daily fish market located in Chioggia (Venice, Italy). Both technologies showed excellent accuracy (100%) in discriminating fishing origin between the Adriatic Sea and Atlantic Ocean. However, the genetic information was not sufficient to differentiate cuttlefishes between the closest fishing areas within the Adriatic Sea (balanced accuracy = 50–83.5%) and did not provide information about the production system (TFP certification), while it showed the highest performance of genetic membership assignment (≥96%) among untreated cuttlefishes collected in the fish market. However, the NIRS model completely differentiated (accuracy = 100%) samples according to the specificity of the treatment nature (simple skinned products vs. TFP preparations), while it was unable to differentiate cuttlefish origin in untreated samples collected in the fish market (accuracy = 50%). The present results demonstrated that NIRS represents a rapid, fast, green and nondestructive approach to support on-site, practical inspection to authenticate and promote TFP protection that mirrors a niche product as a result of the synergy between natural resources and cultural heritage.