Classification Model Research Articles

On the transferability of residence time distributions in two 10-km long river sections with similar hydromorphic units

Quantifying hydrologic exchange fluxes (HEFs) at the stream-groundwater interface and their residence time distributions (RTDs) in the subsurface are important for managing the water quality and ecosystem health in dynamic river corridors. However, field measurements and directly simulating high-spatial resolution HEFs and RTDs can be time-consuming, particularly at watershed-scale. Recent research has proposed the potential application of using hydromorphic units (HUs), which cluster the bathymetry and surface water hydrodynamics attributes, to aid in RTD estimation. However, more pioneering studies have demonstrated that underground structure and large-scale river channel geomorphology are the primary factors controlling the HEFs and RTDs. To address this contradiction, this work evaluates the HEFs and resulting HU-RTD relationships for two 10-km long river sections along the Columbia River, leveraging a one-way coupled three-dimensional transient surface–subsurface water transport modeling framework. Applying such a framework at the two river sections with similar HUs allows for quantitative comparisons of HEFs and RTDs using both statistical tests and machine learning classification models. The comparison reveals that the similarity and transferability of the HU-RTD relationship is very low for the two investigated river sections. This suggests that, creating a general algorithm to estimate RTDs in large-scale river reaches based solely on surface water hydrodynamics and short-distance bathymetry topography data may be nearly impossible.

Advanced differential evolution for gender-aware English speech emotion recognition

Speech emotion recognition (SER) technology involves feature extraction and prediction models. However, recognition efficiency tends to decrease because of gender differences and the large number of extracted features. Consequently, this paper introduces a SER system based on gender. First, gender and emotion features are extracted from speech signals to develop gender recognition and emotion classification models. Second, according to gender differences, distinct emotion recognition models are established for male and female speakers. The gender of speakers is determined before executing the corresponding emotion model. Third, the accuracy of these emotion models is enhanced by utilizing an advanced differential evolution algorithm (ADE) to select optimal features. ADE incorporates new difference vectors, mutation operators, and position learning, which effectively balance global and local searches. A new position repairing method is proposed to address gender differences. Finally, experiments on four English datasets demonstrate that ADE is superior to comparison algorithms in recognition accuracy, recall, precision, F1-score, the number of used features and execution time. The findings highlight the significance of gender in refining emotion models, while mel-frequency cepstral coefficients are important factors in gender differences.

An improved autoencoder for denoising acoustic emission signals in rock fracturing

ABSTRACT Rock fracture acoustic emission (AE) signals are commonly used non-destructive testing data in geological exploration, resource exploitation, and engineering fields. However, these signals are often accompanied by noise interference caused by environmental factors. In this study, we propose an enhanced model for denoising rock fracture AE signals, called simplified fully convolutional denoising autoencoder (SFCDAE). This model is based on the denoising autoencoder principle in the field of deep learning neural networks. The SFCDAE model consists of only seven layers, with minimal preprocessing of data input. By comparing denoising performance evaluation indicators, higher peak signal-to-noise ratio (PSNR) and lower root mean square error (RMSE) were achieved. On average, PSNR increased by 5.575% and RMSE decreased by 22.225%. Using simulated environmental noise to validate the model, it was found that the model has good robustness and can remove artefacts from sudden noise. The practical application value of the LSTM classification model was validated using data containing real experimental noise, resulting in a higher classification accuracy of 80.083%. These results indicate that the proposed model has better denoising performance compared to existing intelligent models and has certain practical value.

Impact of Tile Size and Tile Overlap on the Prediction Performance of Convolutional Neural Networks Trained for Road Classification

Popular geo-computer vision works make use of aerial imagery, with sizes ranging from 64 × 64 to 1024 × 1024 pixels without any overlap, although the learning process of deep learning models can be affected by the reduced semantic context or the lack of information near the image boundaries. In this work, the impact of three tile sizes (256 × 256, 512 × 512, and 1024 × 1024 pixels) and two overlap levels (no overlap and 12.5% overlap) on the performance of road classification models was statistically evaluated. For this, two convolutional neural networks used in various tasks of geospatial object extraction were trained (using the same hyperparameters) on a large dataset (containing aerial image data covering 8650 km2 of the Spanish territory that was labelled with binary road information) under twelve different scenarios, with each scenario featuring a different combination of tile size and overlap. To assess their generalisation capacity, the performance of all resulting models was evaluated on data from novel areas covering approximately 825 km2. The performance metrics obtained were analysed using appropriate descriptive and inferential statistical techniques to evaluate the impact of distinct levels of the fixed factors (tile size, tile overlap, and neural network architecture) on them. Statistical tests were applied to study the main and interaction effects of the fixed factors on the performance. A significance level of 0.05 was applied to all the null hypothesis tests. The results were highly significant for the main effects (p-values lower than 0.001), while the two-way and three-way interaction effects among them had different levels of significance. The results indicate that the training of road classification models on images with a higher tile size (more semantic context) and a higher amount of tile overlap (additional border context and continuity) significantly impacts their performance. The best model was trained on a dataset featuring tiles with a size of 1024 × 1024 pixels and a 12.5% overlap, and achieved a loss value of 0.0984, an F1 score of 0.8728, and an ROC-AUC score of 0.9766, together with an error rate of 3.5% on the test set.

PREDICTION OF PROSPECTIVE NEW STUDENTS USING DECISION TREE, RANDOM FOREST, AND NAIVE BAYES

Higher education positions new student enrollment as a strategic activity for private universities. The effectiveness of selecting prospective students with a high potential to register and be accepted is crucial. Therefore, this study was conducted to find a data classification model that can determine the potential acceptance of new students, allowing private universities to increase the number of students admitted. This research's data originated from the 2020 new student admissions at a prominent private university in Pontianak city. The three chosen classification methods are the decision tree, random forest, and naïve-bayes. Evaluation results indicate the accuracy rate of the decision tree is 59.1%, random forest at 59.2%, and naïve-Bayes at 58.1%. Despite similar accuracy rates, the random forest method slightly outperformed the others, suggesting it may be the most reliable for predicting student enrollment. Based on these models, the estimated potential of prospective students registering at the university ranges from 72% to 78% of the total student candidates. In conclusion, although the three models have almost similar accuracy rates, all show an optimistic estimate regarding the registration potential of prospective students. Thus, universities can use one or a combination of the three models to enhance efficiency in the student admission process.

On Measurement Validity and Language Models: Increasing Validity and Decreasing Bias with Instructions

ABSTRACT Language models like BERT or GPT are becoming increasingly popular measurement tools, but are the measurements they produce valid? Literature suggests that there is still a relevant gap between the ambitions of computational text analysis methods and the validity of their outputs. One prominent threat to validity is hidden biases in the training data, where models learn group-specific language patterns instead of the concept researchers want to measure. This paper investigates to what extent these biases impact the validity of measurements created with language models. We conduct a comparative analysis across nine group types in four datasets with three types of classification models, focusing on the robustness of models against biases and on the validity of their outputs. While we find that all types of models learn biases, the effects on validity are surprisingly small. In particular when models receive instructions as an additional input, they become more robust against biases from the fine-tuning data and produce more valid measurements across different groups. An instruction-based model (BERT-NLI) sees its average test-set performance decrease by only 0.4% F1 macro when trained on biased data and its error probability on groups it has not seen during training increases only by 0.8%.

Network Intrusion Detection and Classification System: A Supervised Machine Learning Approach

Abstract: Intrusion detection systems (IDSs) are crucial for computer security, as they identify and counteract malicious activities within computer networks. Anomaly-based IDSs, specifically, use classification models trained on historical data to detect these harmful activities. This paper proposes an enhanced IDS based on 3-level training and testing of machine learning models, feature selection, resampling, and normalization using Decision Tree, Gaussian Naïve Bayes, K-Nearest Neighbours, Logistic Regression, Random Forest, and Support Vector Machine. In the first stage, the six models are trained and evaluated using the original datasets after pre-processing. In the second stage, the models are built and tested with a resampled version of the dataset using the Synthetic Minority Oversampling Technique (SMOTE). In the third stage, the models are trained and tested with a dataset that has been both resampled and normalized using the standard scaling method. We employ the feature importance technique using the random forest model to select the essential features from NSL-KDD and UNSW-NB15 datasets. The results of our study surpass previous related research, with the decision tree achieving an accuracy, precision, recall, and F1 score of 99.99% on the UNSW-NB15 dataset. Additionally, the decision tree recorded an accuracy of 99.98%, precision of 99.97%, recall of 99.97%, and F1 score of 99.99% on the NSL-KDD dataset.

Application of Spatial Offset Raman Spectroscopy (SORS) and Machine Learning for Sugar Syrup Adulteration Detection in UK Honey

Honey authentication is a complex process which traditionally requires costly and time-consuming analytical techniques not readily available to the producers. This study aimed to develop non-invasive sensor methods coupled with a multivariate data analysis to detect the type and percentage of exogenous sugar adulteration in UK honeys. Through-container spatial offset Raman spectroscopy (SORS) was employed on 17 different types of natural honeys produced in the UK over a season. These samples were then spiked with rice and sugar beet syrups at the levels of 10%, 20%, 30%, and 50% w/w. The data acquired were used to construct prediction models for 14 types of honey with similar Raman fingerprints using different algorithms, namely PLS-DA, XGBoost, and Random Forest, with the aim to detect the level of adulteration per type of sugar syrup. The best-performing algorithm for classification was Random Forest, with only 1% of the pure honeys misclassified as adulterated and <3.5% of adulterated honey samples misclassified as pure. Random Forest was further employed to create a classification model which successfully classified samples according to the type of adulterant (rice or sugar beet) and the adulteration level. In addition, SORS spectra were collected from 27 samples of heather honey (24 Calluna vulgaris and 3 Erica cinerea) produced in the UK and corresponding subsamples spiked with high fructose sugar cane syrup, and an exploratory data analysis with PCA and a classification with Random Forest were performed, both showing clear separation between the pure and adulterated samples at medium (40%) and high (60%) adulteration levels and a 90% success at low adulteration levels (20%). The results of this study demonstrate the potential of SORS in combination with machine learning to be applied for the authentication of honey samples and the detection of exogenous sugars in the form of sugar syrups. A major advantage of the SORS technique is that it is a rapid, non-invasive method deployable in the field with potential application at all stages of the supply chain.

Detecting elevated left ventricular end diastolic pressure from simultaneously measured femoral pressure waveform and electrocardiogram.

Instantaneous, non-invasive evaluation of left ventricular end-diastolic pressure (LVEDP) would have significant value in the diagnosis and treatment of heart failure. A new approach called cardiac triangle mapping (CTM) has been recently proposed, which can provide a non-invasive estimate of LVEDP. We hypothesized that a hybrid machine-learning (ML) method based on CTM can instantaneously identify an elevated LVEDP using simultaneously measured femoral pressure waveform and electrocardiogram (ECG).

Approach: We studied 46 patients (Age: 39-90 (66.4±9.9), BMI: 20.2-36.8 (27.6±4.1), 12 females) scheduled for clinical left heart catheterizations or coronary angiograms at USC Keck Medical Center. Exclusion criteria included severe mitral/aortic valve disease; severe carotid stenosis; aortic abnormalities; ventricular paced rhythm; left bundle branch and anterior fascicular blocks; interventricular conduction delay; and atrial fibrillation. Invasive LVEDP and pressure waveforms at the iliac bifurcation were measured using transducer-tipped Millar catheters with simultaneous ECG. LVEDP range was 9.3-40.5 mmHg. LVEDP=18 mmHg was used as cutoff. Random forest classifiers were trained using data from 36 patients and blindly tested on 10 patients. 

Main Results: Our proposed ML classifier models accurately predict true LVEDP classes using appropriate physics-based features, where the most accurate demonstrates 100.0% (elevated) and 80.0% (normal) success in predicting true LVEDP classes on blind data. 

Significance: We demonstrated that physics-based ML models can instantaneously classify LVEDP using information from femoral waveforms and ECGs. Although an invasive validation, the required ML inputs can be potentially obtained non-invasively.

Novel biomarkers reveal mismatch between tissue and serum thyroid hormone status in amiodarone-induced hyperthyroidism.

Serum TSH and thyroid hormone (TH) levels are routine markers of thyroid function. However, their diagnostic performance is limited under special conditions, e.g. in amiodarone-induced hyperthyroidism (AIH). Such cases would require the assessment of tissue TH action, which is currently unfeasible. Development of an approach that determines how well serum parameters are reflected in tissue TH action of patients. TH-responsive marker genes were identified from human hair follicles (HF) with Next Generation Sequencing, validated by qPCR. A classification model was built with these markers to assess tissue TH action and was deployed on amiodarone treated patients. The impact of amiodarone on tissue TH action was also studied in Thyroid Hormone Action Indicator (THAI) mice. The classification model was validated and shown to predict tissue TH status of subjects with good performance. Serum- and HF-based TH statuses were concordant in hypothyroid and euthyroid amiodarone treated patients. In contrast, amiodarone decreased the coincidence of serum-based and HF-based TH statuses in hyperthyroid patients, indicating that AIH is not unequivocally associated with tissue hyperthyroidism. This was confirmed in the THAI model, where amiodarone prevented tissue hyperthyroidism in THAI mice despite high serum fT4. We developed a minimally-invasive approach using HF markers to assess tissue TH economy that could complement routine diagnostics in controversial cases. We observed that a substantial proportion of AIH patients do not develop tissue hyperthyroidism, indicating that amiodarone protects tissues from thyrotoxicosis. Assessing tissue TH action in patients with AIH may be warranted for treatment decisions.

Machine Learning Predicts Peripherally Inserted Central Catheters-Related Deep Vein Thrombosis Using Patient Features and Catheterization Technology Features.

This study aims to use patient feature and catheterization technology feature variables to train the corresponding machine learning (ML) models to predict peripherally inserted central catheters-deep vein thrombosis (PICCs-DVT) and analyze the importance of the two types of features to PICCs-DVT from the aspect of "input-output" correlation. To comprehensively and systematically summarize the variables used to describe patient features and catheterization technical features, this study combined 18 literature involving the two types of features in predicting PICCs-DVT. A total of 21 variables used to describe the two types of features were summarized, and feature values were extracted from the data of 1,065 PICCs patients from January 1, 2021 to August 31, 2022, to construct a data sample set. Then, 70% of the sample set is used for model training and hyperparameter optimization, and 30% of the sample set is used for PICCs-DVT prediction and feature importance analysis of three common ML classification models (i.e. support vector classifier [SVC], random forest [RF], and artificial neural network [ANN]). In terms of prediction performance, this study selected four metrics to evaluate the prediction performance of the model: precision (P), recall (R), accuracy (ACC), and area under the curve (AUC). In terms of feature importance analysis, this study chooses a single feature analysis method based on the "input-output" sensitivity principle-Permutation Importance. For the mean model performance, the three ML models on the test set are P = 0.92, R = 0.95, ACC = 0.88, and AUC = 0.81. Specifically, the RF model is P = 0.95, R = 0.96, ACC = 0.92, AUC = 0.86; the ANN model is P = 0.92, R = 0.95, ACC = 0.88, AUC = 0.81; the SVC model is P = 0.88, R = 0.94, ACC = 0.85, AUC = 0.77. For feature importance analysis, Catheter-to-vein rate (RF: 91.55%, ANN: 82.25%, SVC: 87.71%), Zubrod-ECOG-WHO score (RF: 66.35%, ANN: 82.25%, SVC: 44.35%), and insertion attempt (RF: 44.35%, ANN: 37.65%, SVC: 65.80%) all occupy the top three in the ML models prediction task of PICCs-DVT, showing relatively consistent ranking results. The ML models show good performance in predicting PICCs-DVT and reveal a relatively consistent ranking of feature importance from the data. The important features revealed might help clinical medical staff to better understand and analyze the formation mechanism of PICCs-DVT from a data-driven perspective.

Artificial Intelligence for Climate Change Biology: From Data Collection to Predictions.

In the era of big data, ecological research is experiencing a transformative shift, yet advancements in thermal ecology and the study of animal responses to climate conditions remain limited. This review discusses how big data analytics and artificial intelligence (AI) can significantly enhance our understanding of microclimates and animal behaviors under changing climatic conditions. We explore AI's potential to refine microclimate models and analyze data from advanced sensors and camera technologies, which capture detailed, high-resolution information. This integration allows researchers to dissect complex ecological and physiological processes with unprecedented precision. We describe how AI can enhance microclimate modeling through improved bias correction and downscaling techniques, providing more accurate estimates of the conditions that animals face under various climate scenarios. Additionally, we explore AI's capabilities in tracking animal responses to these conditions, particularly through innovative classification models that utilize sensors such as accelerometers and acoustic loggers. Moreover, the widespread usage of camera traps can benefit from AI-driven image classification models to accurately identify thermoregulatory responses, such as shade usage and panting. AI is therefore instrumental in monitoring how animals interact with their environments, offering vital insights into their adaptive behaviors. Finally, we discuss how these advanced data-driven approaches can inform and enhance conservation strategies. Detailed mapping of microhabitats essential for species survival under adverse conditions can guide the design of climate-resilient conservation and restoration programs that prioritize habitat features crucial for biodiversity resilience. In conclusion, the convergence of AI, big data, and ecological science heralds a new era of precision conservation, essential for addressing the global environmental challenges of the 21st century.

Chinese traditional painting style automatic classification based on dual-channel feature fusion with multi-attention mechanism

Existing classification models for traditional Chinese paintings mostly ignore shallow detail features, which leads to the imprecise classification of styles. To address the above problems, this paper proposes a Chinese traditional painting style automatic classification model based on dual-channel feature fusion with multi-attention mechanism. First, the spatial attention mechanism is introduced to enhance the Swin-Transformer framework to obtain the salient features of Chinese ancient painting images. Second, a dual-channel attention mechanism is constructed to extract global semantic features and local features of Chinese ancient painting images. Finally, the extracted features are fused and categorized based on the softmax classifier. To verify the feasibility and validity of the proposed model, this paper performs simulations on the Chinese painting dataset and compares it with existing algorithms.The average classification accuracy of the proposed model is 90.6[Formula: see text], with an improvement of 3.14[Formula: see text], which is better than the existing model in both visual effects and objective data comparisons.

Simulation data-driven fault diagnosis method for metro traction motor bearings under small samples and missing fault samples

Abstract Traction motor bearings are crucial for guaranteeing the safe operation of metro vehicles. However, in the metro traction motor bearing fault diagnosis, there are usually problems of small samples and missing fault samples, leading to inaccurate results. Therefore, a novel bearing fault diagnosis method utilizing a track-vehicle-bearing coupled dynamic model and the improved deep convolutional generative adversarial network-multiscale convolutional neural network with mixed attention (IDCG-MAMCNN) model is proposed in this paper. The IDCG-MAMCNN model combines an improved deep convolutional generative adversarial network (IDCGAN) with a multi-scale convolutional neural network with mixed attention (MA-MCNN). Specifically, simulation data is first provided by the coupled dynamic model to supplement missing fault samples. Secondly, the IDCGAN, along with a training method that involves pre-training models with simulation samples and fine-tuning models with experimental samples, is introduced to generate high-quality samples and augment experimental samples under small samples. Lastly, the MA-MCNN serves as the classification model, trained with the augmented dataset comprising experimental, simulation, and generated samples. The fault diagnosis performance of the proposed method is evaluated on the experimental samples of two bearing datasets under small samples and various conditions of missing fault samples. It has been demonstrated by the experimental results that the proposed method exhibits robust fault diagnosis performance and generates high-quality samples under small samples and missing fault samples. Furthermore, the proposed method showcases its adaptability to different operation speeds.

Soil Organic Carbon Prediction Based on Vis–NIR Spectral Classification Data Using GWPCA–FCM Algorithm

Soil visible and near–infrared reflectance spectroscopy is an effective tool for the rapid estimation of soil organic carbon (SOC). The development of spectroscopic technology has increased the application of spectral libraries for SOC research. However, the direct application of spectral libraries for SOC prediction remains challenging due to the high variability in soil types and soil–forming factors. This study aims to address this challenge by improving SOC prediction accuracy through spectral classification. We utilized the European Land Use and Cover Area frame Survey (LUCAS) large–scale spectral library and employed a geographically weighted principal component analysis (GWPCA) combined with a fuzzy c–means (FCM) clustering algorithm to classify the spectra. Subsequently, we used partial least squares regression (PLSR) and the Cubist model for SOC prediction. Additionally, we classified the soil data by land cover types and compared the classification prediction results with those obtained from spectral classification. The results showed that (1) the GWPCA–FCM–Cubist model yielded the best predictions, with an average accuracy of R2 = 0.83 and RPIQ = 2.95, representing improvements of 10.33% and 18.00% in R2 and RPIQ, respectively, compared to unclassified full sample modeling. (2) The accuracy of spectral classification modeling based on GWPCA–FCM was significantly superior to that of land cover type classification modeling. Specifically, there was a 7.64% and 14.22% improvement in R2 and RPIQ, respectively, under PLSR, and a 13.36% and 29.10% improvement in R2 and RPIQ, respectively, under Cubist. (3) Overall, the prediction accuracy of Cubist models was better than that of PLSR models. These findings indicate that the application of GWPCA and FCM clustering in conjunction with the Cubist modeling technique can significantly enhance the prediction accuracy of SOC from large–scale spectral libraries.

Sentiment analysis of passenger feedback on U.S. airlines using machine learning classification methods

Twitter, a platform for micro-blogging, has contained as a novel information architecture. Everyday People worldwide publish about 200 million status messages, known as tweets. Twitter users express their opinions by posting concise text messages. Twitter data is useful for sentiment analysis and consumer feedback tweets. This study employed multi-class sentiment analysis to analyze tweets from 6 major US airlines (American, United, US Airways, Southwest, Delta and Virgin America). Airlines are essential for travel, and this study has helped people choose the best ones. Classification model with the lowest error rate could help airline companies improve their business by figuring out why information is being misclassified. This analysis of airline evaluations can help us identify good airlines and apply this model to our own journeys. This helps the airline identify its weaknesses so they can improve them. A technique of natural language processing (NLP) known as sentiment analysis (or opinion mining) classifies the tone of data as positive, negative, or neutral. The analysis was conducted with seven distinct classification strategies: Linear Discriminant Analysis, Quadratic Discriminant Analysis, Decision Tree, Random Forest, K-Nearest Neighbors, Gradient Boosting, and AdaBoost to utilize the split validation (80% as train data set, 20% as test data set) and 10 folds cross validation process. The suggested model demonstrates superior accuracy and efficiency compared to all others, achieving an accuracy score of 90.13% for the Random Forest with 10 folds cross validation approach. The project aims to utilize machine learning techniques to estimate the reasons for misclassified information since the lowest error rate means the airline sentiment provides less wrong prediction.

Automated detection of abnormal general movements from pressure and positional information in hospitalized infants.

Prechtl's general movements assessment (GMA) allows visual recognition of movement patterns that, when abnormal (cramped synchronized, or CS), have very high sensitivity in predicting later neuromotor disorders; however, training requirements and subjective perceptions from some clinicians may hinder universal adoption of the GMA in the newborn period. To address this, we used a three-phased approach to design a preliminary and clinically-oriented approach to automated CS GMA detection. 335 hospitalized infants were dually recorded on video and a pressure-sensor mat that collected time, spatial, and pressure data. Video recordings were scored by advanced GMA readers. We then conducted a series of unsupervised machine learning and supervised classification modeling with features extracted from clinician- and mat-driven datasets. Finally, the resulting algorithm was converted to a software interface. A classification model combining normalization, clustering, and decision tree modeling resulted in the highest sensitivity for CS movements (100%). Results were delivered via the software interface within 20 min of data recording. The combination of clinical research, machine learning, and repurposing of existing sensor mat technology produced a feasible preliminary approach to automatically detect abnormal GMA in infants while still in the NICU. Further refinements of software and algorithms are needed. Machine learning can differentiate cramped synchronized general movement patterns in the neonatal intensive care unit with good sensitivity and specificity. Increasing access to the GMA through automated detection methods may allow for earlier identification of a greater number of children at high risk for movement delay. Large studies leveraging new artificial intelligence approaches could increase the impact of such detection.

An epilepsy classification based on FFT and fully convolutional neural network nested LSTM

Background and objectiveEpilepsy, which is associated with neuronal damage and functional decline, typically presents patients with numerous challenges in their daily lives. An early diagnosis plays a crucial role in managing the condition and alleviating the patients’ suffering. Electroencephalogram (EEG)-based approaches are commonly employed for diagnosing epilepsy due to their effectiveness and non-invasiveness. In this study, a classification method is proposed that use fast Fourier Transform (FFT) extraction in conjunction with convolutional neural networks (CNN) and long short-term memory (LSTM) models.MethodsMost methods use traditional frameworks to classify epilepsy, we propose a new approach to this problem by extracting features from the source data and then feeding them into a network for training and recognition. It preprocesses the source data into training and validation data and then uses CNN and LSTM to classify the style of the data.ResultsUpon analyzing a public test dataset, the top-performing features in the fully CNN nested LSTM model for epilepsy classification are FFT features among three types of features. Notably, all conducted experiments yielded high accuracy rates, with values exceeding 96% for accuracy, 93% for sensitivity, and 96% for specificity. These results are further benchmarked against current methodologies, showcasing consistent and robust performance across all trials. Our approach consistently achieves an accuracy rate surpassing 97.00%, with values ranging from 97.95 to 99.83% in individual experiments. Particularly noteworthy is the superior accuracy of our method in the AB versus (vs.) CDE comparison, registering at 99.06%.ConclusionOur method exhibits precise classification abilities distinguishing between epileptic and non-epileptic individuals, irrespective of whether the participant’s eyes are closed or open. Furthermore, our technique shows remarkable performance in effectively categorizing epilepsy type, distinguishing between epileptic ictal and interictal states versus non-epileptic conditions. An inherent advantage of our automated classification approach is its capability to disregard EEG data acquired during states of eye closure or eye-opening. Such innovation holds promise for real-world applications, potentially aiding medical professionals in diagnosing epilepsy more efficiently.

Stacked ensemble learning based on deep transfer learning models for food ingredient classification and food quality determination

Stacked ensemble learning based on deep transfer learning models for food ingredient classification and food quality determination

Predicting Tracheostomy Need on Admission to the Intensive Care Unit-A Multicenter Machine Learning Analysis.

It is difficult to predict which mechanically ventilated patients will ultimately require a tracheostomy which further predisposes them to unnecessary spontaneous breathing trials, additional time on the ventilator, increased costs, and further ventilation-related complications such as subglottic stenosis. In this study, we aimed to develop a machine learning tool to predict which patients need a tracheostomy at the onset of admission to the intensive care unit (ICU). Retrospective Cohort Study. Multicenter Study of 335 Intensive Care Units between 2014 and 2015. The eICU Collaborative Research Database (eICU-CRD) was utilized to obtain the patient cohort. Inclusion criteria included: (1) Age >18 years and (2) ICU admission requiring mechanical ventilation. The primary outcome of interest included tracheostomy assessed via a binary classification model. Models included logistic regression (LR), random forest (RF), and Extreme Gradient Boosting (XGBoost). Of 38,508 invasively mechanically ventilated patients, 1605 patients underwent a tracheostomy. The XGBoost, RF, and LR models had fair performances at an AUROC 0.794, 0.780, and 0.775 respectively. Limiting the XGBoost model to 20 features out of 331, a minimal reduction in performance was observed with an AUROC of 0.778. Using Shapley Additive Explanations, the top features were an admission diagnosis of pneumonia or sepsis and comorbidity of chronic respiratory failure. Our machine learning model accurately predicts the probability that a patient will eventually require a tracheostomy upon ICU admission, and upon prospective validation, we have the potential to institute earlier interventions and reduce the complications of prolonged ventilation.