Mean Absolute Percentage Error Research Articles

Predicting microplastic quantities in Indonesian provincial rivers using machine learning models.

Microplastic pollution has surfaced as a critical environmental concern, affecting ecosystems and human health globally. This study explored the application of several machine learning models, including the Tree algorithm, k-Nearest Neighbors (kNN), Random Forest (RF), Linear Regression (LR), Support Vector Machine (SVM), and Neural Networks (NN), to predict microplastic concentrations in the rivers of Indonesia's 24 provinces. By utilizing both environmental and anthropogenic data, the Tree algorithm exhibited the best performance, achieving a coefficient of determination (R2) of 0.838 and a mean absolute percentage error (MAPE) of 0.242 on unseen testing data, thereby highlighting strong predictive capability. Key variables influencing microplastic abundance included annual average temperature, gross domestic product (GDP) per capita and population density. The results underscored the necessity of utilizing comprehensive datasets for effective modeling and highlighted the potential of machine learning to enhance environmental monitoring efforts. This research provides critical insights for policymakers and stakeholders aiming to address the growing issue of microplastic pollution in freshwater systems, providing a foundation for the development of more effective environmental management strategies.

Machine-learning-based cost prediction models for inpatients with mental disorders in China

BackgroundMental disorders are increasingly prevalent, leading to increased medical expenditures. To refine the reimbursement of medical costs for inpatients with mental disorders by health insurance, an accurate prediction model is essential. Per-diem payment is a common internationally implemented payment method for medical insurance of inpatients with mental disorders, necessitating the exploration of advanced machine learning methods for predicting the average daily hospitalization costs (ADHC) based on the characteristics of inpatients with mental disorders.MethodsWe used data including demographic information, clinical/functional characteristics, institutional features, and cost information of 5070 hospitalized patients with mental disorders in Jinhua, China, and employed six algorithms to predict ADHC. Performance of these six algorithms was evaluated through 5- old cross-validation combined with bootstrap method to select the most suitable algorithm and identify key factors influencing ADHC.ResultsThe random forest (RF) model demonstrated better performance (R-squared (R2) = 0.6417 (95% CI, 0.6236–0.6611), root-mean-square error (RMSE) = 0.2398 (95% CI, 0.2252–0.2553), mean-absolute error (MAE) = 0.1677 (95% CI, 0.1626–0.1735), mean-absolute-percentage error (MAPE) = 0.0295 (95% CI, 0.0287–0.0304)). According to feature importance ranking, models incorporating top 11 factors (> 0.01) demonstrated comparable performance to those encompassing all variables. Top four factors (> 0.05) were level of medical institution, age, functional classification, and cognitive classification. Notably, level of medical institutions was the most significant factor across all primary models. Higher medical institutions level, patients below 20 and above 75 years old, lower functional classification, and lower cognitive classification are associated with increased ADHC.ConclusionsMachine learning algorithms, particularly RF algorithm, enhance accuracy of predicting ADHC for mental health patients. The findings of this study provide evidence for setting up more reasonable insurance payment standards for inpatients with mental disorders and support resource allocation in clinical practice.

Reliability and Accuracy of the Fitbit Charge 4 Photoplethysmography Heart Rate Sensor in Ecological Conditions: Validation Study.

Wrist-worn photoplethysmography (PPG) sensors allow for continuous heart rate (HR) measurement without the inconveniences of wearing a chest belt. Although green light PPG technology reduces HR measurement motion artifacts, only a limited number of studies have investigated the reliability and accuracy of wearables in non-laboratory-controlled conditions with actual specific and various physical activity movements. The purpose of this study was to (1) assess the reliability and accuracy of the PPG-based HR sensor of the Fitbit Charge 4 (FC4) in ecological conditions and (2) quantify the potential variability caused by the nature of activities. We collected HR data from participants who performed badminton, tennis, orienteering running, running, cycling, and soccer while simultaneously wearing the FC4 and the Polar H10 chest belt (criterion sensor). Skin tone was assessed with the Fitzpatrick Skin Scale. Once data from the FC4 and criterion data were synchronized, accuracy and reliability analyses were performed, using intraclass correlation coefficients (ICCs), Lin concordance correlation coefficients (CCCs), mean absolute percentage errors (MAPEs), and Bland-Altman tests. A linear univariate model was also used to evaluate the effect of skin tone on bias. All analyses were stratified by activity and pooled activity types (racket sports and running sports). A total of 77.5 hours of HR recordings from 26 participants (age: mean 21.1, SD 5.8 years) were analyzed. The highest reliability was found for running sports, with ICCs and CCCs of 0.90 and 0.99 for running and 0.80 and 0.93 for orienteering running, respectively, whereas the ICCs and CCCs were 0.37 and 0.78, 0.42 and 0.88, 0.65 and 0.97, and 0.49 and 0.81 for badminton, tennis, cycling, and soccer, respectively. We found the highest accuracy for running (bias: 0.1 beats per minute [bpm]; MAPE 1.2%, SD 4.6%) and the lowest for badminton (bias: -16.5 bpm; MAPE 16.2%, SD 14.4%) and soccer (bias: -16.5 bpm; MAPE 17.5%, SD 20.8%). Limit of agreement (LOA) width and artifact rate followed the same trend. No effect of skin tone was observed on bias. LOA width, bias, and MAPE results found for racket sports and soccer suggest a high sensitivity to motion artifacts for activities that involve "sharp" and random arm movements. In this study, we did not measure arm motion, which limits our results. However, whereas individuals might benefit from using the FC4 for casual training in aerobic sports, we cannot recommend the use of the FC4 for specific purposes requiring high reliability and accuracy, such as research purposes.

IoT-Enabled Methane Monitoring and LSTM-Based Forecasting System for Enhanced Safety in Underground Coal Mining

Ensuring safety in the mining industry is a critical concern for a nation's industrial advancement. Industry 4.0, characterized by the integration of advanced technologies, is at the forefront of efforts to enhance mining practices. Coal seams contain a range of hydrocarbon gases, predominantly methane, which is released in significant quantities during mining operations. Effectively mitigating methane emissions is imperative. The inclusion of methane forecasting allows for the early identification of potential methane emissions, hence resulting in significance enhancement in mine safety. The research work is focused on real-time remote monitoring and cloud-based forecasting of methane levels in underground coal mines. An Industrial Internet of Things (IIoT) device is developed for data acquisition in underground coal mines, capturing essential parameters such as methane concentration, temperature, and humidity. The collected data are utilized to train a long short-term memory based multivariate forecasting model. The trained model is subsequently deployed in the cloud. The experiment is performed in a mine of Eastern Coalfields Limited, India. After the deployment of the proposed model, the developed IIoT device transmits real-time data, obtained from the mine, to the cloud. Based on the real-time data, our model conducts methane forecasting and communicates results back to the IIoT device. The device issues immediate alerts when methane levels surpass predefined thresholds. This ensures enhanced safety in mining operations by providing warnings for both current and forecasted methane concentrations. The forecasted methane concentrations, along with real-time data, are accessible through mobile applications and a web-based dashboard. The accuracy of the proposed model is measured by mean absolute error, mean absolute percentage error, and root mean square error, which demonstrate values of 156.95 ppm, 4.23%, and 191.53 ppm, respectively. A comparative study is performed where our model is evaluated against the multivariate multilayer perceptron, vector autoregression, and auto-regressive integrated moving average models. The comparative study demonstrates that our developed model outperforms the others, showing superior results.

Short-Term Load Forecasting Model Based on Time Series Clustering and Transformer in Smart Grid

Accurate Short-Term Load Forecasting (STLF) is a critical task in managing and operating smart grids. Existing STLF methods primarily rely on mathematical modeling or neural networks, often struggle to effectively capture the correlations between influencing factors and load data, and frequently lack interpretability. To address these challenges, this paper proposes an intelligent framework for STLF that combines a pattern extraction and attention mechanism, which leverages the characteristics of electricity consumption data. The proposed framework facilitates the integration of prior knowledge, identifies intrinsic data patterns, and more accurately maps the relationships between influencing factors and load patterns. Finally, we conduct experiments on real-world and publicly available datasets to evaluate the performance of the proposed model. Specifically, the proposed model improves the accuracy of STLF compared to that of existing methods and reduces the mean absolute percentage error by 2% to 5%. The model performs superiorly on the real datasets, with root mean squared error and mean absolute percentage error values of 0.810 MWh and 7.09%.

Estimating Age and Sex from Dental Panoramic Radiographs Using Neural Networks and Vision–Language Models

Purpose: The purpose of this study was to compare multiple deep learning models for estimating age and sex using dental panoramic radiographs and identify the most successful deep learning models for the specified tasks. Methods: The dataset of 437 panoramic radiographs was divided into training, validation, and testing sets. Random oversampling was used to balance the class distributions in the training data and address the class imbalance in sex and age. The models studied were neural network models (CNN, VGG16, VGG19, ResNet50, ResNet101, ResNet152, MobileNet, DenseNet121, DenseNet169) and vision–language models (Vision Transformer and Moondream2). Binary classification models were built for sex classification, while regression models were developed for age estimations. Sex classification was evaluated using precision, recall, F1 score, accuracy, area under the curve (AUC), and a confusion matrix. For age regression, performance was evaluated using mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE), R2, and mean absolute percentage error (MAPE). Results: In sex classification, neural networks achieved accuracies of 85% and an AUC of 0.85, while Moondream2 had much lower accuracy (49%) and AUC (0.48). DenseNet169 performed better than other models for age regression, with an R2 of 0.57 and an MAE of 7.07. Among sex classes, the CNN model achieved the highest precision, recall, and F1 score for both males and females. Vision Transformers that specialised in identifying objects from images demonstrated weaker performance in dental panoramic radiographs, with an inference time of 4.5 s per image. Conclusions: The CNN and DenseNet169 were the most effective models for classifying sex and age regression, performing better than other models for estimating age and sex from dental panoramic radiographs.

Machine Learning-Based Lithium Battery State of Health Prediction Research

To address the problem of predicting the state of health (SOH) of lithium-ion batteries, this study develops three models optimized using the particle swarm optimization (PSO) algorithm, including the long short-term memory (LSTM) network, convolutional neural network (CNN), and support vector regression (SVR), for accurate SOH estimation. Key features were extracted by analyzing the temperature, voltage, and current curves of the battery, and health factors with high correlation to SOH were selected as model inputs using the Pearson correlation coefficient. The PSO algorithm was employed to optimize model parameters, resulting in the construction of three predictive models: PSO-LSTM, PSO-CNN, and PSO-SVR. The models were validated using the NASA PCoE battery aging datasets B0005, B0006, and B0007, with prediction accuracy evaluated based on Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination (R2). Results indicate that the optimized models achieved significant improvements in prediction accuracy, with RMSE and MAE reduced by over 0.5%, a minimum reduction of 38% in MAPE, and R2 exceeding 0.8, demonstrating strong fitting capabilities and validating the effectiveness of the PSO strategy. Among the three models, PSO-LSTM exhibited the best predictive performance, achieving a minimum MAE of 0.67%, RMSE of 0.94%, MAPE of 45.82%, and R2 as high as 0.9298 across the three datasets. These findings suggest that the PSO-LSTM model provides a robust reference for accurate SOH prediction of lithium-ion batteries and shows promising potential for practical applications.

The accuracy of forecasted hospital admission for respiratory tract infections in children aged 0-5 years for 2017/2023.

Healthcare services are in need of tools that can help to ensure a sufficient capacity in periods with high prevalence of respiratory tract infections (RTIs). During the COVID-19 pandemic, we forecasted the number of hospital admissions for RTIs among children aged 0-5 years. Now, in 2024, we aim to examine the accuracy and usefulness of our forecast models. We conducted a retrospective analysis using data from 753,070 children aged 0-5 years, plotting the observed monthly number of RTI admissions, including influenza coded RTI, respiratory syncytial virus (RSV) coded RTI, COVID-19 coded RTI, and other upper and lower RTI, from January 1st, 2017, until May 31st, 2023. We determined the accuracy of four different forecast models, all based on monthly hospital admissions and different assumptions regarding the pattern of virus transmission, computed with ordinary least squares regression adjusting for seasonal trends. We compared the observed vs. forecasted numbers of RTIs between October 31st, 2021, and May 31st, 2023, using metrics such as mean absolute error (MAE), mean absolute percentage error (MAPE) and dynamic time warping (DTW). In our most accurate prediction, we assumed that the proportion of children who remained uninfected and non-hospitalized during the lockdown would be prone to hospitalization in the subsequent season, resulting in increased numbers when lockdown measures were eased. In this prediction, the difference between observed and forecasted numbers at the peak of hospitalizations requiring vs. not requiring respiratory support in November 2021 to January 2022 was 26 (394 vs. 420) vs. 48 (1810 vs. 1762). In scenarios similar to the COVID-19 pandemic, when the transmission of respiratory viruses is suppressed for an extended period, a simple regression model, assuming that non-hospitalized children would be hospitalized the following season, most accurately forecasted hospital admission numbers. These simple forecasts may be useful for capacity planning activities in hospitals.

The Validation of a Novel, Sex-Specific LDL-Cholesterol Equation and the Friedewald, Sampson-NIH, and Extended-Martin-Hopkins Equations Against Direct Measurement in Korean Adults.

Background/Objectives: The currently established equations for calculating low-density lipoprotein cholesterol (LDLc) do not reflect the sex-specific differences in lipid metabolism. We aimed to develop a sex-specific LDLc equation (SSLE) and validate it with three established equations (Friedewald, Sampson-NIH, and ext-Martin-Hopkins) against direct LDLc measurement in Korean adults. Methods: This study included 23,757 subjects (51% male; median age, 51 years) from the 2009-2022 Korean National Health and Nutrition Examination Survey. We developed the SSLE through multiple linear regression incorporating total cholesterol (TC), high-density lipoprotein cholesterol (HDLc), triglycerides (TG), and sex. The validation metrics included Bland-Altman analysis for mean absolute percentage error (MAPE) and agreement of the categorization based on the NCEP ATP-III guidelines, assessed by sex and lipid subgroups. Results: The derived SSLE equation was as follows: for TG < 200 mg/dL, LDLc = 0.963 × TC - 0.881 × HDLc - 0.111 × TG + 0.982 × Sex - 6.958; for TG ≥ 200 mg/dL, LDLc = 0.884 × TC - 0.646 × HDLc - 0.126 × TG + 3.742 × Sex - 3.214 (male = 1, female = 0). The MAPE was similar between males and females for the SSLE (4.6% for both) and ext-Martin-Hopkins (5.0% vs. 4.9%) but higher in males for the Sampson-NIH (5.4% vs. 4.9%) and Friedewald (7.6% vs. 5.7%). In the TG ≥ 400 mg/dL group, the MAPE increased progressively: SSLE (10.2%), ext-Martin-Hopkins (12.0%), Sampson-NIH (12.7%), and Friedewald (27.4%). In the LDLc < 70 mg/dL group, the MAPE was as follows: SSLE (8.0%), Sampson-NIH (8.6%), ext-Martin-Hopkins (9.7%), and Friedewald (12.8%). At TG 200-400 mg/dL, the SSLE revealed very good agreement (κ = 0.801) versus good agreement for other equations (ext-Martin-Hopkins κ = 0.794, Sampson-NIH κ = 0.782, Friedewald κ = 0.696). Conclusions: The novel SSLE demonstrated superior accuracy and agreement in Korean adults. Further validation studies across different ethnic populations are warranted.

Concurrent validity and accuracy of wrist-wearable devices to track heart rate during exercise in sedentary individuals

Background: Heart rate (HR) is commonly used as an indicator to represent a change in oxygen consumption and energy expenditure during activities. Nowadays, HR is easily measured by low-cost wrist-wearable devices, but a few studies examine the validity and accuracy of these devices in sedentary people. Objective: The study aimed to explore the concurrent validity and accuracy of low-cost wrist-wearable HR devices (The GT2e: approx. 4000 THB and Red Mi watch 2 lite: approx. 1000 THB) for measuring HR during exercise on a treadmill in people with a sedentary lifestyle. Materials and methods: Seventy-six sedentary participants (60 female; 78.95%) were instructed to wear a Polar H7 while randomly wearing a watch on each hand. Participants were asked to walk or run on the treadmill for 40 minutes, including free-living activities, exercise, and HR recovery phases. Pearson’s correlation coefficient was utilized to explain the levels of correlation (concurrent validity) of wrist wearable devices with the Polar H7. In contrast, the Bland-Altman method, Concordance correlation coefficient (CCC), and Mean Absolute Percentage Error (MAPE) were then used to determine the accuracy of wrist wearable devices. Results: The GT2e had an excellent agreement with the Polar H7 in free-living activities, exercise sessions, and the HR recovery phase (CCC=0.88, 0.85 &amp; 0.78, respectively) and strong correlation with the criterion measured; Polar H7 (r=0.79 - 0.88; p&lt;0.001). While Red Mi watch 2 lite also had an excellent agreement and correlation in free-living activities and the HR recovery phase (CCC=0.88 - 0.85; r=0.79), but a moderate agreement was found in the exercise phase (CCC=0.55; r=0.61 [p&gt;&lt;0.001]). Conclusion: Both the GT2e and Red Mi watch 2 lite wrist-wearable devices could be used as alternative HR-measured devices to detect HR in sedentary people’s daily lives. However, the GT2e wrist-wearable device was more valid and accurate in detecting HR than the Red Mi watch 2 lite.

AI-Based Discrimination of Faradaic Current against Nonfaradaic Current Inspired by Speech Denoising.

Cyclic voltammetry (CV) has been a powerful technique to provide impactful insights for electrochemical systems, including reaction mechanism, kinetics, diffusion coefficients, etc., in various fields of study, notably energy storage and energy conversion. However, the separation between the faradaic current component of CV and the nonfaradaic current contribution to extract useful information remains a major issue for researchers. Herein, we report a deep learning algorithm inspired by speech denoising that utilizes the theoretical faradaic current as a study target and predicts it from the overall current response from cyclic voltammograms. This deep neural network (DNN) is constructed from a series of fully connected layers, which apply a weight matrix to the inputs and transform it using an activation function to obtain the desired regression. Our model performed well with overall mean absolute percentage errors (MAPEs) of 6.36% between theoretical faradaic currents and the predicted responses from the total currents, with a peak position difference of 2.56 mV for anodic peaks and 2.44 mV for cathodic ones. Furthermore, the algorithm is also capable of extracting peak current values from experimental data with 3.37% MAPE and minimal peak position error (less than 0.75 mV). This innovative approach may be used as a tool to assist researchers in studying electrochemical systems using CV.

Network-Wide Traffic Volume Estimation Based on Probe Vehicle Data

We propose a method for estimating traffic volumes on all roads of the network, which is relevant to various traffic use cases. By combining large-scale probe-vehicle data with stationary detector data, this method builds a model to estimate the probe-vehicle penetration rate at road level, which then allows for traffic volume estimation. Trained by penetration rates from only 60% of the stationary detectors in the Netherlands, our model is able to predict the probe-vehicle penetration rate with a mean absolute percentage error (MAPE) of 8.3%, which decreases to 6% on motorways. Increasing the proportion of detectors used for training did not significantly improve model performance. This suggests that our method can be used in countries with less-developed detector infrastructure as well, provided a minimum coverage of detectors is achieved for all regions and road classes of interest. Traffic volume estimation quality ([Formula: see text]) is constrained by the quality of the penetration rate estimation ([Formula: see text]) and the sampling error, denoted as [Formula: see text]. Using the estimated penetration rates, traffic volumes on motorways can be estimated with a [Formula: see text] of 25% within about 2 min of collecting probe-vehicle observations and 7% within a day. Thus, our traffic volume estimations can support real-time operations on motorways. For roads other than motorways, the maximum accuracy is limited to a [Formula: see text] of about 12%, and significantly longer observation periods are needed. Still, most offline use cases can be covered (e.g., annual average daily traffic).

Effective Dose Estimation in Computed Tomography by Machine Learning.

Computed tomography scans are widely used in everyday medical practice due to speed, image reliability, and detectability of a wide range of pathologies. Each scan exposes the patient to a radiation dose, and performing a fast estimation of the effective dose (E) is an important step for radiological safety. The aim of this work is to estimate E from patient and CT acquisition parameters in the absence of a dose-tracking software exploiting machine learning. In total, 69,037 CT acquisitions were collected with the dose-tracking software (DTS) available at our institution. E calculated by DTS was chosen as the target value for prediction. Different machine learning algorithms were selected, optimizing parameters to achieve the best performance for each algorithm. Effective dose was also estimated using DLP and k-factors, and with multiple linear regression. Mean absolute error (MAE, mean absolute percentage error (MAPE), and R2 were used to evaluate predictions in the test set and in an external dataset of 3800 acquisitions. The random forest regressor (MAE: 0.416 mSv; MAPE: 7%; and R2: 0.98) showed best performances over the neural network and the support vector machine. However, all three machine learning algorithms outperformed effective dose estimation using k-factors (MAE: 2.06; MAPE: 26%) or multiple linear regression (MAE: 0.98; MAPE: 44.4%). The random forest regressor on the external dataset showed an MAE of 0.215 mSv and an MAPE of 7.1%. Our work demonstrated that machine learning models trained with data calculated by a dose-tracking software can provide good estimates of the effective dose just from patient and scanner parameters, without the need for a Monte Carlo approach.

Geometric series exists in nature: Evidence from sorted area sequences of floral parts and leaves.

The concept of a geometric series (GS) plays an important role in mathematics. However, it has been neglected in describing biological size series. Herein, we show that a GS describes the nonreproductive (perianth) parts of the flowers of four Magnoliaceae species and two Rosaceae species and the leaves of 60 Alangium chinense and 60 Shibataea chinensis shoots. The sorted areas of floral parts and leaves formed a sequence that was fitted by a GS with the mean of the quotients of two adjacent members in the sequence as the common ratio of a GS. The mean absolute percent error (MAPE) was used to measure the goodness of fit of each GS. Over 99.7% of the MAPE values (371 out of the 372 tested flowers) were less than 10%, and over 97.8% of the MAPE values were less than 5%. Likewise, over 77.5% of the MAPE values (93 out of the 120 tested shoots) were less than 10%, and over 35% of the MAPE values were less than 5%. These analyses provide empirical evidence that the GS exists in nature, and confirm the usefulness of a classical algebraic formula for the study of plant developmental biology.

Comparative Study among MAPE, RMSE and R Square over the Treatment Techniques Undergone for PCOS Influenced Women

Background: Among the various statistical measures, Root Mean Square Error (RMSE), Mean absolute Percentage Error (MAPE), and R-squared (Coefficient of determination) are the most widely used methods. The significance of the R square approach in the medical field was extensively discussed in the current review. Furthermore, we compared a number of statistical metrics for potential applications in the treatment of various disorders. In addition, the pertinent patents of R square for the consequences of testosterone and the enzymes aspartate dehydrogenase (AST) and alanine transaminase (ALT) on Polycystic Ovary Syndrome (PCOS) treated patients have been developed. Method: We study in this paper the detailed comparative study on the biological system using RMSE, MAPE, and R Squared, which consists of 29 PCOS-influenced women against 20 healthy women and followed by the obesity verification model over the Sprague Dawley rats. Results: R Square provides the best results among all mathematical regression analytical methods in PCOS-influenced patients. Conclusion: In this study, we provide the strong conclusion that aspartate dehydrogenase (AST) with testosterone treated on PCOS influenced women to have a greater chance of getting affected by Non-alcoholic fatty liver disease (NAFLD) rather than alanine transaminase (ALT) with testosterone- treated patients. Furthermore, this study extends their mathematical regression analysis through R squared for the obesity verification over rat model. It confirms that letrozole-treated rats are inhibited in obese compared with control rats, which results in a chance of NAFLD. Therefore, AST combined with testosterone creates a major chance for liver dysfunction.

Can BlazePose accurately assess joint angles in outdoor running environments?

Abstract Study aim: Accurate measurement of joint angles during running is crucial for performance optimisation and injury prevention. Traditional methods are often expensive or impractical for outdoor settings, making pose estimation techniques a promising alternative. This study aims to validate and evaluate the accuracy of BlazePose for measuring hip, knee, and ankle joint angles at key gait events, initial contact (IC), midstance (MS), and toe-off (TO) in outdoor running. Material and methods: Sixteen male athletes participated in this study. Joint angles were measured using BlazePose and compared to manual measurements obtained with Kinovea software. Statistical analysis, including paired t-tests, Pearson correlation coefficients (r), Bland-Altman plots, mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE), was performed to assess BlazePose’s validity, agreement, and accuracy. Results: BlazePose demonstrated a strong to very strong correlation with Kinovea (r: 0.70-0.97). Bland-Altman plots indicated good precision with minimal bias (average MAE: 2.11°, MAPE: 1.57%, RMSE: 2.63°). Ankle joint measurements were the most accurate across all joints and events, followed by the hip joint, particularly during midstance. Knee joint measurements had slightly higher errors but remained within an acceptable range. Conclusions: BlazePose is a valid tool for measuring joint angles during outdoor running, providing an affordable alternative for biomechanical analysis. The findings are specific to the outdoor conditions under which the study was conducted, highlighting the necessity of validating pose estimation techniques in diverse environments and populations for broader applicability.

The importance of adequate preparation of the ocular surface before biometry in reducing refractive errors in patients undergoing cataract surgery

Aims/Purpose: To evaluate the impact of dry eye disease (DED) on refractive errors in patients after cataract surgery.Methods: Fifty patients undergoing cataract surgery were studied. Two study groups and a control group were distinguished. The first study group (group A, n = 22, 44%) were patients with preoperatively diagnosed DED who received appropriate treatment (eyelid margin hygiene and moisturizing the ocular surface). The second study group (group B, n = 14, 28%) were patients in whom such treatment was also implemented, despite the lack of a diagnosis of DED. The remaining patients (n = 14, 28%) were the control group (Group C). Biometric measurements with calculation of intraocular lens (IOL) power were performed twice in all of them (before and after treatment of DED). Refraction and refractive errors were evaluated one month after surgery.Results: Significant improvement in best corrected visual acuity (BCVA) was obtained in all groups after surgical treatment. In both study groups, a significant reduction in the mean absolute error (MAE) after the treatment was achieved – in study group A, the MAE was 0.37 ± 0.30 D vs 0.24 ± 0.33 D, p = 0.003. In study group B, MAE was 0.29 ± 0.26 D vs 0.23 ± 0.23 D, p = 0.017. The control group showed no difference in MAE (0.27 ± 0.18D vs 0.33 ± 0.21 D, p = 0.322). In study group A, there was a significant reduction in the percentage of MAE in the ± 0.25 D range (36.36% vs 77.27%, p = 0.016). The results are comparable with previous publications.Conclusions: Diagnosis and treatment of dry eye syndrome before cataract surgery is an important factor in reducing refractive errors after surgery. The use of lubricating drops and eyelid margin hygiene before biometry can be considered for all patients undergoing cataract surgery.

The importance of adequate preparation of the ocular surface before biometry in reducing refractive errors in patients undergoing cataract surgery

Aims/Purpose: To evaluate the impact of dry eye disease (DED) on refractive errors in patients after cataract surgery.Methods: Fifty patients undergoing cataract surgery were studied. Two study groups and a control group were distinguished. The first study group (group A, n = 22, 44%) were patients with preoperatively diagnosed DED who received appropriate treatment (eyelid margin hygiene and moisturizing the ocular surface). The second study group (group B, n = 14, 28%) were patients in whom such treatment was also implemented, despite the lack of a diagnosis of DED. The remaining patients (n = 14, 28%) were the control group (Group C). Biometric measurements with calculation of intraocular lens (IOL) power were performed twice in all of them (before and after treatment of DED). Refraction and refractive errors were evaluated one month after surgery.Results: Significant improvement in best corrected visual acuity (BCVA) was obtained in all groups after surgical treatment. In both study groups, a significant reduction in the mean absolute error (MAE) after the treatment was achieved – in study group A, the MAE was 0.37 ± 0.30 D vs 0.24 ± 0.33 D, p = 0.003. In study group B, MAE was 0.29 ± 0.26 D vs 0.23 ± 0.23 D, p = 0.017. The control group showed no difference in MAE (0.27 ± 0.18D vs 0.33 ± 0.21 D, p = 0.322). In study group A, there was a significant reduction in the percentage of MAE in the ± 0.25 D range (36.36% vs 77.27%, p = 0.016). The results are comparable with previous publications.Conclusions: Diagnosis and treatment of dry eye syndrome before cataract surgery is an important factor in reducing refractive errors after surgery. The use of lubricating drops and eyelid margin hygiene before biometry can be considered for all patients undergoing cataract surgery.

Stacking Ensemble Learning Method for Quantitative Analysis of Soluble Solid Content in Apples

ABSTRACTThe soluble solids content (SSC) in apples directly affects their quality. This study aimed to detect SSC nondestructively using hyperspectral technology combined with chemometrics. However, data generation may not follow a specific pattern, and even small perturbations in the data can have a significant impact on the constructed model. To improve the anti‐interference capability of individual models, this study proposed a stacking ensemble learning method that adopted partial least squares (PLS), support vector machine (SVM), extreme gradient boosting (Xgboost), random forest (RF) as basic‐learners, and RF serving as a meta‐learner. Experimental results showed that the performance of the established model on the test set were as follows: the root mean square error (RMSE) was 0.4325, mean absolute error (MAE) was 0.3245, mean absolute percentage error (MAPE) was 0.0271, coefficient of determination () was 0.9250. These results indicate that the stacking ensemble learning approach could appropriately fuse the predictive results of each basic‐learner and improve the prediction accuracy of individual models. To verify the superiority of the proposed stacking ensemble learning method, the selection of its basic‐learners, meta‐learner, and combination strategy were compared and analyzed. This study not only provides a theoretical reference for the further development of related nondestructive detection equipment but also offers guidance for fusion algorithms as well.

A Deep Learning–Based Model for Real-Time Production Forecasting of Gas Wells with Intermittent Production

Summary Accurate real-time production forecasting for gas wells experiencing intermittent production due to liquid loading is critical for optimizing operating schedules. Traditional methods, which rely on solving wellbore multiphase flow equations, are computationally intensive, making them impractical for real-time production forecasting across hundreds or thousands of wells, especially in the absence of reliable downhole measurements. This study presents an efficient real-time prediction model for production parameters of intermittent gas wells using deep learning techniques. The model generates production forecasts based on high-frequency measurements, including casing and tubing pressures, valve status, and instantaneous gas rates. To enhance the model’s performance, a novel cycle parameter extraction procedure is introduced during the feature engineering phase. This procedure extracts key performance indicators, specifically production parameters at critical moments, based on the shut-in and open cycles. Using these key cycle parameters as features significantly reduces the data volume used for model training and enhances prediction accuracy. The cycle parameter data set is derived from high-frequency measurements collected from 216 low-productivity gas wells. The developed deep learning model uses a Seq2Seq architecture. The encoding long short-term memory (LSTM) processes recent production data and passes the encoded information to the decoding LSTM, which then uses future operating schedules as inputs to predict the corresponding key cycle parameters. Testing results demonstrate the model’s high accuracy and efficiency, with each prediction taking less than 0.002 seconds. The prediction accuracy is quantified using the coefficient of determination (R²), the mean absolute error (MAE), and the mean absolute percentage error (MAPE). The R² values for the predicted features range from 91% to 99%, with a maximum MAE of 0.15 MPa for pressure parameters and 0.10×104 m³/d for the average gas rate (AGR). Moreover, the model maintains high accuracy even in challenging scenarios, such as wells with extended shut-in durations and insufficient measurements. The model’s predictive performance under various operating schedules is further evaluated using virtual cycles, demonstrating its strong generalization capability in generating production forecasts for unseen operating schedules in a given well. This generalization capability underscores the advantage of training the model with data collected from numerous wells operating under diverse schedules.