Learning Predictive Models Research Articles

Development of machine learning models for fractional flow reserve prediction in angiographically intermediate coronary lesions.

Fractional flow reserve (FFR) represents the gold standard in guiding the decision to proceed or not with coronary revascularization of angiographically intermediate coronary lesion (AICL). Optical coherence tomography (OCT) allows to carefully characterize coronary plaque morphology and lumen dimensions. We sought to develop machine learning (ML) models based on clinical, angiographic and OCT variables for predicting FFR. Data from a multicenter, international, pooled analysis of individual patient's level data from published studies assessing FFR and OCT on the same target AICL were collected through a dedicated database to train (n = 351) and validate (n = 151) six two-class supervised ML models employing 25 clinical, angiographic and OCT variables. A total of 502 coronary lesions in 489 patients were included. The AUC of the six ML models ranged from 0.71 to 0.78, whereas the measured F1 score was from 0.70 to 0.75. The ML algorithms showed moderate sensitivity (range: 0.68-0.77) and specificity (range: 0.59-0.69) in detecting patients with a positive or negative FFR. In the sensitivity analysis, using 0.75 as FFR cut-off, we found a higher AUC (0.78-0.86) and a similar F1 score (range: 0.63-0.76). Specifically, the six ML models showed a higher specificity (0.71-0.84), with a similar sensitivity (0.58-0.80) with respect to 0.80 cut-off. ML algorithms derived from clinical, angiographic, and OCT parameters can identify patients with a positive or negative FFR.

Machine Learning Models for Gold Price Prediction: A Comparative Analysis and Evaluation

The prediction of gold prices is meaningful for multiple industries. Gold is a relatively stable store of value, and its price is closely linked to stocks, exchange rates, and monetary policy. Observing the trend of gold prices can promote the stable development of financial businesses. In this article, the daily opening and closing prices of gold over the past ten years were chosen as the data support. Seven different models were trained to predict prices on a yearly basis with a half-year interval. These predictions were then compared to actual data, and evaluations and analyses were conducted to identify the most suitable model for gold price forecasting. The machine learning models used included Ridge Regression, Linear Regression, Decision Tree Random Forest, Lasso Regression, Support Vector Machine, and K-Nearest Neighbors. In the evaluation process, Mean Squared Error (MSE) was used as the criterion to assess the accuracy of the model predictions.

Deep Learning Model for Detection and Severity Analysis of Car Accidents

Abstract On-road car accidents are immensely unfortunate but quite common occurrences worldwide. Instant data-centric and informed decisions of crisis management are rarely experienced due to the absence of real-time car accident detection and severity analysis mechanisms. On this background, the current paper presents a deep learning model for car accident detection and analysis of its severity so that the crisis management activities might follow without any delay saving invaluable human lives. The existing works lack in using time-series data, the proper learning model for accurate prediction, and minimizing the time taken in post-accident scenarios for the victims to receive immediate medical help. This paper introduces the Long Short Term Memory (LSTM) model in conjunction with the Gradient Boosted Regression Trees (GBRT) technique for the determination of car accidents with different levels of severity. The proposed model works with the accelerometer and gyroscopic data collected through an application installed in the smartphones of the users inside the car. The LSTM-GBRT hybrid model is proposed to achieve higher accuracy than LSTM which deals with time-variant data. The satisfactory performance of the proposed technique has been reported and the results are extensively investigated in comparison with another hybrid technique such as LSTM with Random Forest (RF) as well. The statistics confirm the superiority of the proposed model over other parallel models in terms of several performance metrics, like Accuracy, Precision, etc.

Fast flow field prediction based on E(2)-equivariant steerable convolutional neural networks

In the field of flow field reconstruction, traditional deep learning models predominantly rely on standard convolutions, but their predictive accuracy remains limited. To address this issue, we explore the potential of E(2)-equivariant convolutions to enhance the predictive accuracy of deep learning models for fast flow field prediction. Unlike conventional convolutions, E(2)-equivariant convolutions offer a richer representation capability by better capturing geometric and structural information. Our neural network integrates an attention mechanism that leverages the signed distance function (SDF) to encode geometric details and an indicator matrix to incorporate boundary conditions. The model predicts velocity and pressure fields as outputs. We conducted experiments specifically targeting non-uniform steady laminar flows, and the results show a 16.1% reduction in overall error compared to models based on traditional convolutions while maintaining high efficiency. These findings indicate that E(2)-equivariant convolution, coupled with an attention mechanism, significantly improves flow field prediction by focusing on critical information and better representing complex geometries.

Multiple Machine Learning Models in House Price Prediction: Performance Evaluation and Comparison

House pricing is one of the most critical concerns for people. Accurately predicting house prices is essential for participants and investors in the real estate market. This study delves into the realm of predictive modeling for house prices using a multimodal approach in machine learning. The research methodology spans various stages including data collection, preprocessing, feature engineering, and model selection. Based on a dataset sourced from Kaggle, housing data from Iowa comprising 79 pertinent features influencing property prices is utilized. Performance evaluation is conducted using the root mean square error (RMSE), comparing the efficacy of linear regression models such as Lasso regression, Kernel Ridge regression, and ElasticNet regression against tree-based models including XGBoost, Gradient Boosting, and LightGBM. Additionally, the study implements two stacked ensemble models to identify the most optimal predictive model for house prices. Experimental procedures involve training and evaluating these models on specific datasets to gauge their predictive accuracy. Findings and discussions underscore the comparative analysis of diverse machine learning models within the context of house price prediction. Moreover, the study employs the Gradient Boosting method in machine learning to pinpoint the ten most influential features impacting house prices. Recommendations for future research are provided to enhance the precision and robustness of house price prediction models. This research contributes to the advancement of predictive modeling in the real estate domain, offering insights into the efficacy of various machine learning techniques for forecasting property prices.

HEnsem_DTIs: A heterogeneous ensemble learning model for drug-target interactions prediction

Drug discovery is the process by which a drug is discovered. Drug-target interactions prediction is a major part of drug discovery. Unfortunately, producing new drugs is time-consuming and expensive; Because it requires a lot of human and laboratory resources. Recently, predictions have been made using computational methods to solve these problems and prevent blindly examining all interactions. Various experiences using computational methods show that no single algorithm can be suitable for all applications; Hence, ensemble learning is expressed. Although various ensemble methods have been proposed, it is still not easy to find a suitable ensemble method for a particular dataset. In general, the existing algorithms in aggregation and combination method are selected manually based on experience. Reinforcement learning can be one way to meet this challenge. High-dimensional feature space and class imbalance are among the challenges of drug-target interactions prediction. This paper proposes HEnsem_DTIs, a heterogeneous ensemble model, for predicting drug-target interactions using dimensionality reduction and concepts of recommender systems to address these challenges. HEnsem_DTIs is configured with reinforcement learning. Dimensionality reduction is applied to handle the challenge of high-dimensional feature space and recommender systems to improve under-sampling and solve the class imbalance challenge. Six datasets are used to evaluate the proposed model; Results of the evaluation on datasets show that HEnsem_DTIs works better than other models in this field. Results of evaluation of the proposed model on the first dataset using 10-fold cross-validation experiments show the amount of sensitivity 0.896, specificity 0.954, GM 0.924, AUC 0.930 and AUPR 0.935.

DMSS: An Attention-Based Deep Learning Model for High-Quality Mass Spectrometry Prediction

DMSS: An Attention-Based Deep Learning Model for High-Quality Mass Spectrometry Prediction

A physics-informed machine learning model for global-local stress prediction of open holes with finite-width effects in composite structures

Fast and accurate methods are required to predict stresses in the vicinity of open and closed holes in composite structures, especially in a global-local modelling context as applied during the design of airframe structures. Fast analytical solutions for infinite-width anisotropic plates with open holes do not consider finite-width effects. Heuristic methods and semi-analytical solutions can be used to towards addressing such effects. To improve the accuracy and speed of these respective methods, we use machine learning (ML) methods trained on high-fidelity finite element analyses to make finite-width corrections. However, such methods require large amounts of training data to reduce errors to satisfactory levels. Therefore, in this study, the fusion of analytical solutions with machine learning is performed. We develop an analytical solution-informed ML model that is as fast as an analytical solution and superior in accuracy to analytical solutions with heuristic finite-width scaling. Our informed ML model offers accuracies equal to analytical solutions for the infinite-width case, and it is capable for use in a global-local modelling context, under uniaxial and biaxial loading. Our informed ML model outperforms prediction accuracy across all cases compared to uninformed ML models and requires a significantly lower size training dataset size.

A novel case of honeycomb shaped pin fin heat sink: CFD-data driven machine learning models for thermal performance prediction

This research explores a useful thermal engineering application utilizing novel honeycomb-shaped pin fins heat sink (PFHS). 3D incompressible flow and heat transfer are examined using standard k-ԑ turbulence model for Reynolds numbers (Re) ranging from 8547 to 21367. Using the PFHS with 1.5 mm side length and pitch arrangement of (S =25 × 25), the Nusselt Number (Nu) increased by 171 % at a Re of 21367. Furthermore, Cu-Diamond composite material raises Hydrothermal performance factor (HTPF) to 3.304. At a Re of 8547, honeycomb heat sink has 200 % greater HTPF than the baseline. Predictive modeling of HTPF, Nu, and pressure drop (ΔP) was done using Artificial Neural Network (ANN), Multiple Linear Regression (MLR), Extreme Gradient Boosting Regression (XGBR), and Light Gradient Boosting Machine Regression (LGBMR). The four models with the highest R2 and lowest MRE on the test dataset performed best. Both models were implemented using Keras and Sklearn. XGBR and MLR have superior HTPF prediction accuracy (R2test = 0.977, MRE (%) = 1.1 and 0.924, MRE (%) = 2.1). XGBR and MLR predicted the Nu well with R2test values of 0.979 and MRE percentages of 1.9 and 3.9. R2test of 0.999 and MRE (%) of 0.3 indicate that XGBR predicts pressure reduction.

A DEEP LEARNING ENSEMBLE APPROACH FOR X-RAY IMAGE CLASSIFICATION

The application of deep learning-based intelligent systems for X-ray imaging in various settings, including transportation, customs inspections, and public security, to identify hidden or prohibited objects are discussed in this study. In busy environments, x-ray inspections face challenges due to time limitations and a lack of qualified personnel. Deep learning algorithms can automate the imaging process, enhancing object detection and improving safety. This study uses a dataset of 5094 x-ray images of laptops with hidden foreign circuits and normal ones, training 11 deep learning algorithms with the 10-fold cross-validation method. The predictions of deep learning models selected based on the 70% threshold value have been combined using a meta-learner. ShuffleNet has the highest individual performance with 83.56%, followed by InceptionV3 at 81.30%, Darknet19 at 78.92%, DenseNet201 at 77.70% and Xception at 71.26%. Combining these models into an ensemble achieved a remarkable classification success rate of 85.97%, exceeding the performance of any individual model. The ensemble learning approach provides a more stable prediction output, reducing standard deviation among folds as well. This research highlights the potential for safer and more effective X-ray inspections through advanced machine learning techniques.

Development and Validation of a Deep Learning Model for Prediction of Adult Physiological Deterioration.

Prediction-based strategies for physiologic deterioration offer the potential for earlier clinical interventions that improve patient outcomes. Current strategies are limited because they operate on inconsistent definitions of deterioration, attempt to dichotomize a dynamic and progressive phenomenon, and offer poor performance. Can a deep learning deterioration prediction model (Deep Learning Enhanced Triage and Emergency Response for Inpatient Optimization [DETERIO]) based on a consensus definition of deterioration (the Adult Inpatient Decompensation Event [AIDE] criteria) and that approaches deterioration as a state "value-estimation" problem outperform a commercially available deterioration score? The derivation cohort contained retrospective patient data collected from both inpatient services (inpatient) and emergency departments (EDs) of two hospitals within the University of California San Diego Health System. There were 330,729 total patients; 71,735 were inpatient and 258,994 were ED. Of these data, 20% were randomly sampled as a retrospective "testing set." The validation cohort contained temporal patient data. There were 65,898 total patients; 13,750 were inpatient and 52,148 were ED. DETERIO was developed and validated on these data, using the AIDE criteria to generate a composite score. DETERIO's architecture builds upon previous work. DETERIO's prediction performance up to 12 hours before T0 was compared against Epic Deterioration Index (EDI). In the retrospective testing set, DETERIO's area under the receiver operating characteristic curve (AUC) was 0.797 and 0.874 for inpatient and ED subsets, respectively. In the temporal validation cohort, the corresponding AUC were 0.775 and 0.856, respectively. DETERIO outperformed EDI in the inpatient validation cohort (AUC, 0.775 vs. 0.721; p < 0.01) while maintaining superior sensitivity and a comparable rate of false alarms (sensitivity, 45.50% vs. 30.00%; positive predictive value, 20.50% vs. 16.11%). DETERIO demonstrates promise in the viability of a state value-estimation approach for predicting adult physiologic deterioration. It may outperform EDI while offering additional clinical utility in triage and clinician interaction with prediction confidence and explanations. Additional studies are needed to assess generalizability and real-world clinical impact.

Portfolio Optimization Strategies: New Approaches Based on Machine Learning Forecasting

This study provides an in-depth discussion and comprehensive review of the latest applications of machine learning techniques in the field of portfolio optimization. The article begins with an overview of traditional portfolio optimization theory and its limitations, and then focuses on how machine learning predictive models, which have flourished in recent years, can provide new perspectives and tools for solving the problems of non-linearity, dynamics and uncertainty in investment decision-making. This paper provides a detailed overview of the application practices of various machine learning algorithms (e.g., deep learning, reinforcement learning, integrated learning, etc.) in the areas of asset return prediction, risk assessment, and optimal weight allocation, and analyses their advantages and challenges compared to traditional methods. The analysis of relevant research cases reveals the significant effect of machine learning predictions in increasing expected portfolio returns, reducing risk exposure, and achieving effective diversification. The study also explores possible future trends and potential research directions for machine learning in portfolio optimization, highlighting the importance of combining domain knowledge with big data-driven intelligent investment decisions. This review aims to provide financial scholars and practitioners with a new way of thinking about portfolio optimization, and to promote the combination of theoretical research and practical operation in the fields of financial engineering and investment management, so as to achieve more accurate and efficient investment decisions.

Predictive Value of Simulated CT Radiomics Combined with Ipsilateral Lung Dosimetry Parameters for Radiation Pneumonitis in Patients with Esophageal Cancer: A Machine Learning-Based Retrospective Study.

To explore how non-surgical esophageal cancer patients can identify high-risk factors for radiation-induced pneumonitis after receiving radiotherapy. We retrospectively included 228 esophageal cancer patients who were unable to undergo surgical treatment but received radiotherapy for the first time. By retrospective analysis and identifying potential risk factors for symptomatic radiation-induced pneumonitis (ie ≥grade 2), as well as delineating the affected lung as an area of interest on localized CT and extracting radiomics features, along with extracting dosimetric parameters from the affected lung area. After feature screening, patients were randomly divided into training and testing sets in a 7-to-3 ratio, and a prediction model was established using machine learning algorithms. Finally, the receiver operating characteristic (ROC) curve and decision curve analysis (DCA) were used to validate the predictive performance of the model. A total of 54 cases of symptomatic radiation pneumonitis occurred in this study, with a total incidence rate of 23.68%. The results of multivariate analysis showed that the occurrence of symptomatic radiation pneumonitis was significantly correlated with the mean lung dose (MLD), esophageal PTVD90, esophageal PTVV50, V5, V10, V15, and V20 in patients. The machine learning prediction model constructed based on candidate prediction variables has a prediction performance interval between 0.751 (95% CI: 0.700-0.802) and 0.891 (95% CI: 0.840-0.942) in the training and validation sets, respectively. Among them, the RFM algorithm has the best prediction performance for radiation-induced pneumonitis, with 0.891 (95% CI: 0.840-0.942) and 0.887 (95% CI: 0.836-0.938) in the training and validation sets, respectively. The combination of localization CT radiomics features and diseased lung dosimetry parameters has good predictive value for radiation-induced pneumonitis in esophageal cancer patients after radiotherapy. Especially, the radiation-induced pneumonitis prediction model constructed using RF algorithm can be more effectively used to guide clinical decision-making in esophageal cancer patients.

Prediction of transfusion risk after total knee arthroplasty: use of a machine learning algorithm

IntroductionTotal knee arthroplasty (TKA) carries a significant hemorrhagic risk, with a non-negligible rate of postoperative transfusions. The blood-sparing strategy has evolved to reduce blood loss after TKA by identifying the patient's risk factors preoperatively. In practice, a blood count is often performed postoperatively but rarely altering the patient's subsequent management. This study aimed to identify the preoperative variables associated with hemorrhagic risk, enabling the creation of a machine-learning model predictive of transfusion risk after total knee arthroplasty and the need for a complete blood count. HypothesisBased on preoperative data, a powerful machine learning predictive model can be constructed to estimate the risk of transfusion after total knee arthroplasty. Material and methodsThis retrospective single-centre study included 774 total knee arthroplasties (TKA) operated between January 2020 and March 2023. Twenty-five preoperative variables were integrated into the machine learning model and filtered by a recursive feature elimination algorithm. The most predictive variables were selected and used to construct a gradient-boosting machine algorithm to define the overall postoperative transfusion risk model. Two groups were formed of patients transfused and not transfused after TKA. Odds ratios were determined, and the area under the curve evaluated the model's performance. ResultsOf the 774 TKA surgery patients, 100 were transfused postoperatively (12.9%). The machine learning predictive model included five variables: age, body mass index, tranexamic acid administration, preoperative hemoglobin level, and platelet count. The overall performance was good with an area under the curve of 0.97 [95% CI 0.921–1], sensitivity of 94.4% [95% CI 91.2–97.6], and specificity of 85.4% [95% CI 80.6–90.2]. The tool developed to assess the risk of blood transfusion after TKA is available at https://arthrorisk.com. ConclusionThe risk of postoperative transfusion after total knee arthroplasty can be predicted by a model that identifies patients at low, moderate, or high risk based on five preoperative variables. This machine learning tool is available on a web platform that is accessible to all, easy to use, and has a high prediction performance. The model aims to limit the need for routine check-ups, depending on the risk presented by the patient. Level of evidenceII; diagnostic study

Predictive modeling of rice milling degree for three typical Chinese rice varieties using interpretative machine learning methods.

Brown rice over-milling causes high economic and nutrient loss. The rice degree of milling (DOM) detection and prediction remain a challenge for moderate processing. In this study, a self-established grain image acquisition platform was built. Degree of bran layer remaining (DOR) datasets is established with image capturing and processing (grain color, texture, and shape features extraction). The mapping relationship between DOR and the DOM is in-depth analyzed. Rice grain DOR typical machine learning and deep learning prediction models are established. The results indicate that the optimized Catboost model can be established with cross-validation and grid search method, with the best accuracy improving from 84.28% to 91.24%, achieving precision 91.31%, recall 90.89%, and F1-score 91.07%. Shapley additive explanations analysis indicates that color, texture, and shape feature affect Catboost prediction accuracy, the feature importance: color>texture>shape. The YCbCr-Cb_ske and GLCM-Contrast features make the most significant contribution to rice milling quality prediction. The feature importance provides theoretical and practical guidancefor grain DOM prediction model. PRACTICAL APPLICATION: Rice milling degree prediction and detection are valuable for rice milling process in practical application. In this paper, image processing and machine learning methods provide an automated, nondestructive, and cost-effective way to predict the quality of rice. The study may serve as a valuable reference for improving rice milling methods, retaining rice nutrition, and reducing broken rice yield.

189P An image-based deep learning prediction model for characterization of the drug tolerant persister cell state

189P An image-based deep learning prediction model for characterization of the drug tolerant persister cell state

P05-06 Conazoles’ health risk prediction: integrating knowledge graphs and machine learning predictive models

P05-06 Conazoles’ health risk prediction: integrating knowledge graphs and machine learning predictive models

Statistical and machine learning models for location-specific crop yield prediction using weather indices.

Crop yield prediction gains growing importance for all stakeholders in agriculture. Since the growth and development of crops are fully connected with many weather factors, it is inevitable to incorporate meteorological information into yield prediction mechanism. The changes in climate-yield relationship are more pronounced at a local level than across relatively large regions. Hence, district or sub-region-level modeling may be an appropriate approach. To obtain a location- and crop-specific model, different models with different functional forms have to be explored. This systematic review aims to discuss research papers related to statistical and machine-learning models commonly used to predict crop yield using weather factors. It was found that Artificial Neural Network (ANN) and Multiple Linear Regression were the most applied models. Support Vector Regression (SVR) model has a high success ratio as it performed well in most of the cases. The optimization options in ANN and SVR models allow us to tune models to specific patterns of association between weather conditions of a location and crop yield. ANN model can be trained using different activation functions with optimized learning rate and number of hidden layer neurons. Similarly, the SVR model can be trained with different kernel functions and various combinations of hyperparameters. Penalized regression models namely, LASSO and Elastic Net are better alternatives to simple linear regression. The nonlinear machine learning models namely, SVR and ANN were found to perform better in most of the cases which indicates there exists a nonlinear complex association between crop yield and weather factors.

Machine Learning Prediction Model to Predict Length of Stay of Patients Undergoing Hip or Knee Arthroplasties: Results from a High-Volume Single-Center Multivariate Analysis.

Background: The growth of arthroplasty procedures requires innovative strategies to reduce inpatients' hospital length of stay (LOS). This study aims to develop a machine learning prediction model that may aid in predicting LOS after hip or knee arthroplasties. Methods: A collection of all the clinical notes of patients who underwent elective primary or revision arthroplasty from 1 January 2019 to 31 December 2019 was performed. The hospitalization was classified as "short LOS" if it was less than or equal to 6 days and "long LOS" if it was greater than 7 days. Clinical data from pre-operative laboratory analysis, vital parameters, and demographic characteristics of patients were screened. Final data were used to train a logistic regression model with the aim of predicting short or long LOS. Results: The final dataset was composed of 1517 patients (795 "long LOS", 722 "short LOS", p = 0.3196) with a total of 1541 hospital admissions (729 "long LOS", 812 "short LOS", p < 0.001). The complete model had a prediction efficacy of 78.99% (AUC 0.7899). Conclusions: Machine learning may facilitate day-by-day clinical practice determination of which patients are suitable for a shorter LOS and which for a longer LOS, in which a cautious approach could be recommended.

Detecting starch-adulterated turmeric using Vis-NIR spectroscopy and multispectral imaging with machine learning

Chemicals are often added to turmeric to increase profits, posing significant health risks to consumers. At the same time, traditional methods for detecting contaminants in turmeric are complicated and time-consuming. This study aimed to develop a more practical approach using visible-near infrared (Vis-NIR) and multispectral imaging (MSI) techniques to detect starch adulteration in turmeric. The turmeric powder was mixed with starch (0.1, 0.5, 1, 2.5, 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, and 100 % (w/w)) to create spectral and MSI datasets within the 400–1050 nm wavelength range. Spectra were corrected using spectral preprocessing techniques such as Savitzky-Golay (SG), Multiplicative Scatter Correction (MSC), and Standard Normal Variate (SNV). Principal Component Analysis (PCA) applied to reduce the dimensions and various machine learning (ML) models for prediction. The Random Forest Regressor (RFR) achieved a coefficient of determination (R²) of 0.999, a root mean squared error (RMSE) of 0.391 mg (w/w), and a residual predictive deviation (RPD) of 92.3 % in regression analysis. For classification, the Random Forest Classifiers (RFC) achieved an F1 score of 96.0 % and a Matthews Correlation Coefficient (MCC) of 94.6 %. In MSI analysis, the DenseNet201 model obtained an F1 score of 92.9 % and an MCC of 91.9 %. Moreover, the robustness of these models was cross-validated using leave-one-out cross-validation (LOOCV) and K-fold methods. The significance of the study lies in several critical areas, such as public health, advancement in technology, etc. The study's findings reveal that Vis-NIR and MSI approaches are excellent in detecting starch adulteration in turmeric with reliability. It has important implications for public health and food safety by offering a reliable tool for verifying the purity of turmeric and other food items.