Performance Of Machine Learning Models Research Articles

Flooded area detection and mapping from Sentinel-1 imagery. Complementary approaches and comparative performance evaluation

ABSTRACT The current study assesses the performance of several machine learning (ML) and deep learning (DL) models for detecting and mapping floods using Sentinel-1 SAR imagery. Three distinct approaches were used: pixel classification, object-based image analysis and object instance segmentation. The ML models are Random Forest, and Support Vector Machine and the DL models are U-NET, DeepLabV3 and Mask RCNN. Five different case studies were selected to test the models’ scalability. These areas are in Romania (Prut River, at the border between Romania, the Republic of Moldova and Ukraine, Timiș River, and Râul Negru River), the United States of America (Missouri River) and Australia (Broughton Creek). For each flood, five Sentinel-1 images were used, four collected before the flood and one collected after the flood. The intensity images were stacked and scaled in the range of the intensity thresholds associated with water and non-water so that all the case studies have the same margins for intensity. Samples of water, vegetation, agricultural fields, and bare soil were collected only from the Prut River case study and used in the training process. Out of all models, the U-Net model returned the highest accuracy with a value for Intersect over Union of 0.763 for a tile size of 128x128 pixels.

Measuring the solubility of tacrolimus in supercritical carbon dioxide (binary and ternary systems), comparing the performance of machine learning models with conventional models

The solubility of tacrolimus in a supercritical carbon dioxide (SC-CO2) with and without co-solvent (ternary and binary system) was investigated. The experimental conditions include the temperatures of 308 to 338 K and the pressures of 12 to 30 MPa were considered. The values of tacrolimus solubilization in SC-CO2 in binary and ternary systems were from 0.416 × 10−5 to 1.589 × 10−5 and from 0.389 × 10−4 to 0.875 × 10−4, respectively. The obtained results presented that adding a co-solvent (ethanol) could affect tacrolimus solubility significantly, the mole fraction of tacrolimus in SC-CO2/co-solvent under all experimental conditions was at least 4.9 more than that of SC-CO2 alone. The correlation of the solubility data was achieved by three distinct methods: (1) SRK EoS with van der Waals (vdW2) and Wong-Sandler (WS) as mixing rules, (2) solution theory, (3) empirical and (4) machine learning based models. The analysis of variance (ANOVA) results showed that Sparks et al. (AIC −817.05, and AARD 5.08 %) and Bian et al. (AIC −815.77 and AARD 5.17 %) models could correlate the experimental data of tacrolimus in SC-CO2 with more accuracy, as compared to the other empirical models. Sodeifian and Sajadian model with AARD% and AIC of 2.99 and −726.22, respectively, has acted as the most accurate empirical model in ternary system. The solution theory was able to calculate the solubility for the tacrolimus-SC-CO2-ethanol system with high accuracy (AARD, % 4.85, AIC −701.41, F value 169.70 and Radj 0.9622). Also, the Gradient Boosting model (AARD accurately predicted the outcomes with perfect accuracy for binary and ternary systems. Finally, total (ΔHtotal, 23.21 cal/mol), solvation (ΔHsol, −19.47 cal/mol) and vaporization (ΔHvap, 42.68 cal/mol) enthalpies were obtained by using an energy term in Chrastil and Bartle models.

GRU-IKF scene taxiing trajectory prediction model based on attention mechanism

Aircraft taxi trajectory prediction helps solve operational problems such as airport taxiing conflicts and long waiting times, ensuring airport safety while improving service levels and increasing airport throughput. In view of the fact that the performance of machine learning models depends on good data sets, a method for predicting taxi trajectories of aircraft on the ground based on attention mechanism, fusion of gated recurrent unit (GRU) and improved Kalman filter algorithm (IKF) is proposed. Firstly, three independent gated recurrent unit networks are used to capture the motion state and temporal dependency of aircraft at future moments, and the attention mechanism is introduced to enhance the ability to extract data difference features and learn the mapping relationship from input to output; then, it is fused with the improved extended Kalman filter to integrate the output results of the neural network into the state prediction and update process to improve the accuracy of the predicted trajectory sequence. Finally, the effectiveness of the model was verified using the actual taxiing trajectory of aircraft at Lukou Airport. The simulation results show that the model can effectively and accurately predict the taxiing trajectory of aircraft on the surface, with an overall mean square error of about 0.00128. Compared with the single recurrent neural network (RNN), long short-term memory network (LSTM) and GRU model, the RMSE is reduced by 72.9%and respectively 39.9%, 54.7%and the prediction time is 40 ms. It can accurately and quickly predict the taxiing trajectory, which helps to reduce the operating load of the airport surface management system.

Machine learning framework for predicting water quality classification

ABSTRACT Groundwater serves as the source for nearly half of the world's drinking water, yet understanding of global groundwater resources remains incomplete, and management of aquifers falls short, particularly concerning groundwater quality. This research offers insights into the groundwater quality in 242 stations of Maharashtra and Union Territory of Dadra and Nagar Haveli and nine parameters (pH, TDS, TH, Calcium (Ca2+), Magnesium (Mg2+), Chloride (Cl−), Sulphate (SO42−), Nitrate (NO3−), Fluoride (F−)) were considered for computing the Water Quality Index (WQI) and hence Water Quality Classification (WQC) based on Water Quality Index (WQI). This research introduces the utilisation of Machine Learning (ML) models, specifically, Random Forest, Adaptive Boosting (AdaBoost), Gradient Boosting, XGBoost, Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) model for predicting WQC and models are tested. Grid search method as a hyperparameter tuning of parameters is utilized to achieve the best possible performance of ML models. The performance metrics that are used for evaluating and reporting the performance of classification models are Accuracy, Precision, Recall or Sensitivity, F1 Score. SVM achieved the highest performance in predicting WQC. With accurate predictions of WQC, these findings have the potential to enhance NEP concerning water resources by facilitating ongoing improvements in water quality.

Simultaneous Recognition and Detection of Adenosine Phosphates by Machine Learning Analysis for Surface-Enhanced Raman Scattering Spectral Data.

Adenosine phosphates (adenosine 5'-monophosphate (AMP), adenosine 5'-diphosphate (ADP), and adenosine 5'-triphosphate (ATP)) play important roles in energy storage and signal transduction in the human body. Thus, a measurement method that simultaneously recognizes and detects adenosine phosphates is necessary to gain insight into complex energy-relevant biological processes. Surface-enhanced Raman scattering (SERS) is a powerful technique for this purpose. However, the similarities in size, charge, and structure of adenosine phosphates (APs) make their simultaneous recognition and detection difficult. Although approaches that combine SERS and machine learning have been studied, they require massive quantities of training data. In this study, limited AP spectral data were obtained using fabricated gold nanostructures for SERS measurements. The training data were created by feature selection and data augmentation after preprocessing the small amount of acquired spectral data. The performances of several machine learning models trained on these generated training data were compared. Multilayer perceptron model successfully detected the presence of AMP, ADP, and ATP with an accuracy of 0.914. Consequently, this study establishes a new measurement system that enables the highly accurate recognition and detection of adenosine phosphates from limited SERS spectral data.

EEG-based emotional valence and emotion regulation classification: a data-centric and explainable approach

Emotion classification using electroencephalographic (EEG) data is a challenging task in the field of Artificial Intelligence. While many researchers have focused on finding the best model or feature extraction technique to achieve optimal results, few have attempted to select the best methodological steps for working with the dataset. In this study, we applied two different theoretical approaches based on the noise of the dataset: curriculum learning and confident learning. Curriculum learning involves presenting training examples to the model in a specific order, starting with easier examples and gradually increasing in difficulty. This approach has been shown to improve model performance. Confident learning is a method for identifying and correcting label errors in datasets. By identifying and correcting these errors, confident learning can improve the performance of machine learning models trained on noisy datasets. We then applied the Integrated Gradient technique in order to assess the explainability of each model. Our aim was to explore the impact of different models and methods on emotion classification performance using EEG data. We collected and used an EEG dataset in which participants rated the emotional valence of positive and negative pictures while performing an emotion regulation (ER) task, comparing a control condition (Look) with two ER strategies: cognitive reappraisal and expressive suppression. We performed a multilabel classification to identify emotional neutrality or polarization of emotional valence (both positive and negative) rated by participants and the emotion regulation strategy adopted during the task. We compared the performance of models trained on three datasets selected based on label noise and evaluated their suitability for this task. Our results suggest different patterns based on the architecture used for feature importance, highlighting both advantages and criticisms.

Comparison of Different Negative-Sample Acquisition Strategies Considering Sample Representation Forms for Debris Flow Susceptibility Mapping

The lack of reliable negative samples is an important factor limiting the quality of machine learning-based debris flow susceptibility mapping (DFSM). The purpose of this paper is to propose multiple negative-sample acquisition strategies for DFSM considering different sample representation forms. The sample representation forms mainly include a single grid, multi-grid, and watershed unit, and the negative-sample acquisition strategies are based on support vector machine (SVM), spy technique, and isolation forest (IF) methods, respectively. These three strategies can assign a value to all the samples based on different assumptions, and reliable, negative samples can be generated from samples with values below a predefined threshold. Combining different sample representation forms with negative sample acquisition strategies, nine datasets were then involved in random forest (RF) modeling. The receiver operating characteristic (ROC) curves and related statistical results were used to evaluate the models. The results show that the strategy based on the spy technique is suitable for multiple datasets, while the IF-based strategy is well-adapted to the watershed unit datasets. This study can provide more options for improving the quality of datasets in DFSM, which can further improve the performance of machine learning models.

Employing machine learning for advanced gap imputation in solar power generation databases

This research evaluates the application of advanced machine learning algorithms, specifically Random Forest and Gradient Boosting, for the imputation of missing data in solar energy generation databases and their impact on the size of green hydrogen production systems. The study demonstrates that the Random Forest model notably excels in harnessing solar data to optimize hydrogen production, achieving superior prediction accuracy with mean absolute error (MAE) of 0.0364, mean squared error (MSE) of 0.0097, root mean squared error (RMSE) of 0.0985, and a coefficient of determination (R2) of 0.9779. These metrics surpass those obtained from baseline models including linear regression and recurrent neural networks, highlighting the potential of accurate imputation to significantly enhance the efficiency and output of renewable energy systems. The findings advocate for the integration of robust data imputation methods in the design and operation of photovoltaic systems, contributing to the reliability and sustainability of energy resource management. Furthermore, this research makes significant contributions by showcasing the comparative performance of traditional machine learning models in handling data gaps, emphasizing the practical implications of data imputation on optimizing hydrogen production systems. By providing a detailed analysis and validation of the imputation models, this work offers valuable insights for future advancements in renewable energy technology.

Improving Predictive Efficacy for Drug Resistance in Novel HIV-1 Protease Inhibitors through Transfer Learning Mechanisms.

The human immunodeficiency virus presents a significant global health challenge due to its rapid mutation and the development of resistance mechanisms against antiretroviral drugs. Recent studies demonstrate the impressive performance of machine learning (ML) and deep learning (DL) models in predicting the drug resistance profile of specific FDA-approved inhibitors. However, generalizing ML and DL models to learn not only from isolates but also from inhibitor representations remains challenging for HIV-1 infection. We propose a novel drug-isolate-fold change (DIF) model framework that aims to predict drug resistance score directly from the protein sequence and inhibitor representation. Various ML and DL models, inhibitor representations, and protein representations were analyzed through realistic validation mechanisms. To enhance the molecular learning capacity of DIF models, we employ a transfer learning approach by pretraining a graph neural network (GNN) model for activity prediction on a data set of 4855 HIV-1 protease inhibitors (PIs). By performing various realistic validation strategies on internal and external genotype-phenotype data sets, we statistically show that the learned representations of inhibitors improve the predictive ability of DIF-based ML and DL models. We achieved an accuracy of 0.802, AUROC of 0.874, and r of 0.727 for the unseen external PIs. By comparing the DIF-based models with a null model consisting of isolate-fold change (IF) architecture, it is observed that the DIF models significantly benefit from molecular representations. Combined results from various testing strategies and statistical tests confirm the effectiveness of DIF models in testing novel PIs for drug resistance in the presence of an isolate.

Predicting 3-month poor functional outcomes of acute ischemic stroke in young patients using machine learning

BackgroundPrediction of short-term outcomes in young patients with acute ischemic stroke (AIS) may assist in making therapy decisions. Machine learning (ML) is increasingly used in healthcare due to its high accuracy. This study aims to use a ML-based predictive model for poor 3-month functional outcomes in young AIS patients and to compare the predictive performance of ML models with the logistic regression model.MethodsWe enrolled AIS patients aged between 18 and 50 years from the Third Chinese National Stroke Registry (CNSR-III), collected between 2015 and 2018. A modified Rankin Scale (mRS) ≥ 3 was a poor functional outcome at 3 months. Four ML tree models were developed: The extreme Gradient Boosting (XGBoost), Light Gradient Boosted Machine (lightGBM), Random Forest (RF), and The Gradient Boosting Decision Trees (GBDT), compared with logistic regression. We assess the model performance based on both discrimination and calibration.ResultsA total of 2268 young patients with a mean age of 44.3 ± 5.5 years were included. Among them, (9%) had poor functional outcomes. The mRS at admission, living alone conditions, and high National Institutes of Health Stroke Scale (NIHSS) at discharge remained independent predictors of poor 3-month outcomes. The best AUC in the test group was XGBoost (AUC = 0.801), followed by GBDT, RF, and lightGBM (AUCs of 0.795, 0, 794, and 0.792, respectively). The XGBoost, RF, and lightGBM models were significantly better than logistic regression (P < 0.05).ConclusionsML outperformed logistic regression, where XGBoost the boost was the best model for predicting poor functional outcomes in young AIS patients. It is important to consider living alone conditions with high severity scores to improve stroke prognosis.

Greylag goose optimization and multilayer perceptron for enhancing lung cancer classification

Lung cancer is an important global health problem, and it is defined by abnormal growth of the cells in the tissues of the lung, mostly leading to significant morbidity and mortality. Its timely identification and correct staging are very important for proper therapy and prognosis. Different computational methods have been used to enhance the precision of lung cancer classification, among which optimization algorithms such as Greylag Goose Optimization (GGO) are employed. These algorithms have the purpose of improving the performance of machine learning models that are presented with a large amount of complex data, selecting the most important features. As per lung cancer classification, data preparation is one of the most important steps, which contains the operations of scaling, normalization, and handling gap factor to ensure reasonable and reliable input data. In this domain, the use of GGO includes refining feature selection, which mainly focuses on enhancing the classification accuracy compared to other binary format optimization algorithms, like bSC, bMVO, bPSO, bWOA, bGWO, and bFOA. The efficiency of the bGGO algorithm in choosing the optimal features for improved classification accuracy is an indicator of the possible application of this method in the field of lung cancer diagnosis. The GGO achieved the highest accuracy with MLP model performance at 98.4%. The feature selection and classification results were assessed using statistical analysis, which utilized the Wilcoxon signed-rank test and ANOVA. The results were also accompanied by a set of graphical illustrations that ensured the adequacy and efficiency of the adopted hybrid method (GGO + MLP).

Integration of FAERS, DrugBank and SIDER Data for Machine Learning-based Detection of Adverse Drug Reactions

AbstractTraditionally, disproportionality analysis (DPA) methods are employed for signal detection in pharmacovigilance, but these methods utilize only a limited portion of the data available from spontaneous event reports (SERs). This research aims to enhance signal detection by applying machine learning (ML) methods that can leverage additional data. We create a dataset by integrating SER data from the FDA Adverse Event Reporting System (FAERS) with biological and chemical data from DrugBank, and information on known adverse drug reactions (ADRs) from Side Effect Resource (SIDER). The known ADRs from SIDER are used to label the dataset for ML training. Using the AutoML library TPOT, ML models are trained on this dataset. Our findings indicate that ML models, even when trained with the same features as DPA methods, achieve higher recall and precision. Moreover, incorporating additional features related to drugs and events significantly boosts the performance of ML models. Analysis using the explainable AI (XAI) technique SHAP reveals that the drug name, event name, and fifth-level ATC code are the most influential features for model predictions. These ML models offer a promising alternative or supplement to conventional DPA methods for signal detection in pharmacovigilance.

Expressibility-induced concentration of quantum neural tangent kernels

Quantum tangent kernel methods provide an efficient approach to analyzing the performance of quantum machine learning models in the infinite-width limit, which is of crucial importance in designing appropriate circuit architectures for certain learning tasks. Recently, they have been adapted to describe the convergence rate of training errors in quantum neural networks in an analytical manner. Here, we study the connections between the expressibility and value concentration of quantum tangent kernel models. In particular, for global loss functions, we rigorously prove that high expressibility of both the global and local quantum encodings can lead to exponential concentration of quantum tangent kernel values to zero. Whereas for local loss functions, such issue of exponential concentration persists owing to the high expressibility, but can be partially mitigated. We further carry out extensive numerical simulations to support our analytical theories. Our discoveries unveil a fundamental feature of quantum neural tangent kernels, indicating that the issue of their concentration cannot be bypassed merely by transitioning to a local encoding scheme while maintaining high expressibility. This offers valuable insights for the design of wide quantum variational circuit models in practical applications.

Feature selection with annealing for forecasting financial time series

Stock market and cryptocurrency forecasting is very important to investors as they aspire to achieve even the slightest improvement to their buy-or-hold strategies so that they may increase profitability. However, obtaining accurate and reliable predictions is challenging, noting that accuracy does not equate to reliability, especially when financial time-series forecasting is applied owing to its complex and chaotic tendencies. To mitigate this complexity, this study provides a comprehensive method for forecasting financial time series based on tactical input–output feature mapping techniques using machine learning (ML) models. During the prediction process, selecting the relevant indicators is vital to obtaining the desired results. In the financial field, limited attention has been paid to this problem with ML solutions. We investigate the use of feature selection with annealing (FSA) for the first time in this field, and we apply the least absolute shrinkage and selection operator (Lasso) method to select the features from more than 1000 candidates obtained from 26 technical classifiers with different periods and lags. Boruta (BOR) feature selection, a wrapper method, is used as a baseline for comparison. Logistic regression (LR), extreme gradient boosting (XGBoost), and long short-term memory are then applied to the selected features for forecasting purposes using 10 different financial datasets containing cryptocurrencies and stocks. The dependent variables consisted of daily logarithmic returns and trends. The mean-squared error for regression, area under the receiver operating characteristic curve, and classification accuracy were used to evaluate model performance, and the statistical significance of the forecasting results was tested using paired t-tests. Experiments indicate that the FSA algorithm increased the performance of ML models, regardless of problem type. The FSA hybrid models showed better performance and outperformed the other BOR models on seven of the 10 datasets for regression and classification. FSA-based models also outperformed Lasso-based models on six of the 10 datasets for regression and four of the 10 datasets for classification. None of the hybrid BOR models outperformed the hybrid FSA models. Lasso-based models, excluding the LR type, were comparable to the best models for six of the 10 datasets for classification. Detailed experimental analysis indicates that the proposed methodology can forecast returns and their movements efficiently and accurately, providing the field with a useful tool for investors.

Single and Multi-Hop Question-Answering Datasets for Reticular Chemistry with GPT-4-Turbo.

The rapid advancement in artificial intelligence and natural language processing has led to the development of large-scale datasets aimed at benchmarking the performance of machine learning models. Herein, we introduce "RetChemQA", a comprehensive benchmark dataset designed to evaluate the capabilities of such models in the domain of reticular chemistry. This dataset includes both single-hop and multi-hop question-answer pairs, encompassing approximately 45,000 question and answers (Q&As) for each type. The questions have been extracted from an extensive corpus of literature containing about 2,530 research papers from publishers including NAS, ACS, RSC, Elsevier, and Nature Publishing Group, among others. The dataset has been generated using OpenAI's GPT-4 Turbo, a cutting-edge model known for its exceptional language understanding and generation capabilities. In addition to the Q&A dataset, we also release a dataset of synthesis conditions extracted from the corpus of literature used in this study. The aim of RetChemQA is to provide a robust platform for the development and evaluation of advanced machine learning algorithms, particularly for the reticular chemistry community. The dataset is structured to reflect the complexities and nuances of real-world scientific discourse, thereby enabling nuanced performance assessments across a variety of tasks.

Comparing Human-Level and Machine Learning ModelPerformance in White Blood Cell MorphologyAssessment.

There is an increasing research focus on the role of machine learning in the haematology laboratory, particularly in blood cell morphologic assessment. Human-level performance is an important baseline and goal for machine learning. This study aims to assess the interobserver variability and human-level performance in blood cell morphologic assessment. A dataset of 1000 single white blood cell images were independently labelled by 10 doctors and morphology scientists. Interobserver variability was calculated using Fleiss' kappa. Observers' labels were then separated into consensus labels used to determine ground truth, and performance labels used to assess observer performance. A machine learning model was trained and assessed using the same cell images. Explainability images (XRAI and IG) were generated for each of the test images. The Fleiss kappa for all 10 observers was 0.608, indicating substantial agreement between observers. The accuracy of human observers was 95%, with sensitivity 72% and specificity 97%. The accuracy of the machine learning model was 95%, with sensitivity 71% and specificity 97%. The model shared similar performance across labels when compared to humans. Explainability metrics demonstrated that the machine learning model was able to differentiate between the cytoplasm and nucleus of the cells, and used these features to perform predictions. The substantial, though not perfect, agreement between human observers highlights the inherent subjectivity in white blood cell morphologic assessment. A machine learning model performed similarly to human observers in single white blood cell identification. Further research is needed to compare human-level and machine learning performance in ways that more closely reflect the typical process of morphologic assessment.

Performance of Machine Learning Models to Predict Customer Satisfaction Scores for Lao National Convention Center

This research paper investigates the effectiveness of various machine learning (ML) models in predicting customer satisfaction scores for the Lao National Convention Center (LNCC). By utilizing a dataset containing customer feedback, demographic data, facilities, and service-related information, the study aims to identify the best performing model for predicting customer satisfaction. In this paper, the performance of ML to predict customer satisfaction scores from the questionnaire survey dataset is evaluated. The customer satisfaction score is categorized into five classes: strongly agree, agree, neutral, disagree and strongly disagree. The analysis will compare the performance of different models, including linear regression (LR), k-nearest neighbor (KNN), support vector machines (SVM), decision trees (DT) and random forest (RF). The results show that SVM model achieved the highest accuracy rate of 95.91%, followed by KNN and LR with an accuracy rate of 95.55% and 95.49%, respectively. The findings of this study have important implications for the use of ML in improving the service-related information for LNCC and providing valuable insights for decision-makers and developers.

An intelligent learning system based on electronic health records for unbiased stroke prediction

Stroke has a negative impact on people’s lives and is one of the leading causes of death and disability worldwide. Early detection of symptoms can significantly help predict stroke and promote a healthy lifestyle. Researchers have developed several methods to predict strokes using machine learning (ML) techniques. However, the proposed systems have suffered from the following two main problems. The first problem is that the machine learning models are biased due to the uneven distribution of classes in the dataset. Recent research has not adequately addressed this problem, and no preventive measures have been taken. Synthetic Minority Oversampling (SMOTE) has been used to remove bias and balance the training of the proposed ML model. The second problem is to solve the problem of lower classification accuracy of machine learning models. We proposed a learning system that combines an autoencoder with a linear discriminant analysis (LDA) model to increase the accuracy of the proposed ML model for stroke prediction. Relevant features are extracted from the feature space using the autoencoder, and the extracted subset is then fed into the LDA model for stroke classification. The hyperparameters of the LDA model are found using a grid search strategy. However, the conventional accuracy metric does not truly reflect the performance of ML models. Therefore, we employed several evaluation metrics to validate the efficiency of the proposed model. Consequently, we evaluated the proposed model’s accuracy, sensitivity, specificity, area under the curve (AUC), and receiver operator characteristic (ROC). The experimental results show that the proposed model achieves a sensitivity and specificity of 98.51% and 97.56%, respectively, with an accuracy of 99.24% and a balanced accuracy of 98.00%.

Pre-operative prediction of BCR-free survival with mRNA variables in prostate cancer.

Technological innovation yielded opportunities to obtain mRNA expression data for prostate cancer (PCa) patients even prior to biopsy, which can be used in a precision medicine approach to treatment decision-making. This can apply in particular to predict the risk of, and time to biochemical recurrence (BCR). Most mRNA-based models currently proposed to this end are designed for risk classification and post-operative prediction. Effective pre-operative prediction would facilitate early treatment decision-making, in particular by indicating more appropriate therapeutic pathways for patient profiles who would likely not benefit from a systematic prostatectomy regime. The aim of this study is to investigate the possibility to leverage mRNA information pre-operatively for BCR-free survival prediction. To do this, we considered time-to-event machine learning (ML) methodologies, rather than classification models at a specific survival horizon. We retrospectively analysed a cohort of 135 patients with clinical follow-up data and mRNA information comprising over 26,000 features (data accessible at NCBI GEO database, accession GSE21032). The performance of ML models including random survival forest, boosted and regularised Cox models were assessed, in terms of model discrimination, calibration, and predictive accuracy for overall, 3-year and 5-year survival, aligning with common clinical endpoints. Results showed that the inclusion of mRNA information could yield a gain in performance for pre-operative BCR prediction. ML-based time-to-event models significantly outperformed reference nomograms that used only routine clinical information with respect to all metrics considered. We believe this is the first study proposing pre-operative transcriptomics models for BCR prediction in PCa. External validation of these findings, including confirmation of the mRNA variables identified as potential key predictors in this study, could pave the way for pre-operative precision nomograms to facilitate timely personalised clinical decision-making.

Resampling approach for imbalanced data classification based on class instance density per feature value intervals

In practical applications, imbalanced datasets significantly degrade the classification performance of machine learning models. However, most conventional resampling approaches fall short in adequately addressing the varying contributions of individual features to the classification model. In response to this defect, this study introduces three novel resampling approaches. The first approach, Oversampling based on class instance density per feature value intervals (OCF), focuses on augmenting the dataset. The second approach, Undersampling based on class instance density per feature value intervals (UCF), seeks to reduce dataset size. The third approach, Hybrid sampling based on class instance density per feature value intervals (HSCF), which can perform oversampling and undersampling simultaneously. These approaches categorize feature value into different intervals based on their varying information content, calculate class instance densities within these intervals, and generate feature values in intervals with high discriminative information. Subsequently, these generated features are combined to synthesize minority class data, effectively achieving oversampling. Additionally, the study combines class instance density and feature importance to identify majority class data at the classification boundary with minimal contribution and subsequently executes undersampling. The flexibility to adjust sampling ratios and the integration of OCF and UCF enable the implementation of hybrid sampling. Finally, experiments on the benchmark dataset demonstrate the superiority and effectiveness of the proposed method. Furthermore, it is observed that the method proposed in this study enhances the feature dividing capability of decision tree classifiers. Hence, the best results are achieved when working in synergy with decision tree classifiers, leading to the most significant improvements in classification performance. All codes have been published at https://github.com/Wangfeiopen/HSCF.