Machine Learning Algorithm Model Research Articles

An External-Validated Prediction Model to Predict Lung Metastasis among Osteosarcoma: A Multicenter Analysis Based on Machine Learning.

Background Lung metastasis greatly affects medical therapeutic strategies in osteosarcoma. This study aimed to develop and validate a clinical prediction model to predict the risk of lung metastasis among osteosarcoma patients based on machine learning (ML) algorithms. Methods We retrospectively collected osteosarcoma patients from the Surveillance Epidemiology and End Results (SEER) database and from four hospitals in China. Six ML algorithms, including logistic regression (LR), gradient boosting machine (GBM), extreme gradient boosting (XGBoost), random forest (RF), decision tree (DT), and multilayer perceptron (MLP), were applied to build predictive models for predicting lung metastasis using patient's demographics, clinical characteristics, and therapeutic variables from the SEER database. The model was internally validated using 10-fold cross-validation to calculate the mean area under the curve (AUC) and the model was externally validated using the Chinese multicenter osteosarcoma data. Relative importance ranking of predictors was plotted to understand the importance of each predictor in different ML algorithms. The correlation heat map of predictors was plotted to understand the correlation of each predictor, selecting the 10-fold cross-validation with the highest AUC value in the external validation ROC curve to build a web calculator. Results Of all enrolled patients from the SEER database, 17.73% (194/1094) developed lung metastasis. The multiple logistic regression analysis showed that sex, N stage, T stage, surgery, and bone metastasis were all independent risk factors for lung metastasis. In predicting lung metastasis, the mean AUCs of the six ML algorithms ranged from 0.711 to 0.738 in internal validation and 0.697 to 0.729 in external validation. Among the six ML algorithms, the extreme gradient boosting (XGBoost) model had the highest AUC value with an average internal AUC of 0.738 and an external AUC of 0.729. The best performing ML algorithm model was used to build a web calculator to facilitate clinicians to calculate the risk of lung metastasis for each patient. Conclusions The XGBoost model may have the best prediction effect and the online calculator based on this model can help doctors to determine the lung metastasis risk of osteosarcoma patients and help to make individualized medical strategies.

Machine learning-driven dynamic risk prediction for highly pathogenic avian influenza at poultry farms in Republic of Korea: Daily risk estimation for individual premises.

Highly pathogenic avian influenza (HPAI) is a fatal zoonotic disease that damages the poultry industry and endangers human lives via exposure to the pathogen. A risk assessment model that precisely predicts high-risk groups and occurrence of HPAI infection is essential for effective biosecurity measures that minimize the socio-economic losses of massive outbreaks. However, the conventional risk prediction approaches have difficulty incorporating the broad range of factors associated with HPAI infections at poultry holdings. Therefore, it is difficult to accommodate the complexity of the dynamic transmission mechanisms and generate risk estimation on a real-time basis. We proposed a continuous risk prediction framework for HPAI occurrences that used machine learning algorithms (MLAs). This integrated environmental, on-farm biosecurity, meteorological, vehicle movement tracking, and HPAI wild bird surveillance data to improve accuracy and timeliness. This framework consisted of (i) the generation of 1788 predictors from six types of data and reconstructed them with an outcome variable into a data mart based on a temporal assumption (i.e. infected period and day-ahead forecasting); (ii) training of the predictors with the temporally rearranged outcome variable that corresponded to HPAI H5N6 infected state at each individual farm on daily basis during the 2016-2017 HPAI epidemic using three different MLAs [Random Forest, Gradient Boosting Machine (GBM), and eXtreme Gradient Boosting]; (iii) predicting the daily risk of HPAI infection during the 2017-2018 HPAI epidemic using the pre-trained MLA models for each farm across the country. The models predicted the high risk to 8-10 out of 19 infected premises during the infected period in advance. The GBM MLAs outperformed the 7-day forecasting of HPAI prediction at individual poultry holdings, with an area under the curve (AUC) of receiver operating characteristic of 0.88. Therefore, this approach enhances the flexibility and timing of interventions against HPAI outbreaks at poultry farms.

Increased risk of COVID-19-related admissions in patients with active solid organ cancer in the West Midlands region of the UK: a retrospective cohort study

ObjectiveSusceptibility of patients with cancer to COVID-19 pneumonitis has been variable. We aim to quantify the risk of hospitalisation in patients with active cancer and use a machine learning algorithm...

Application of Random Forest Algorithm for Merging Multiple Satellite Precipitation Products across South Korea

Precipitation is a crucial component of the water cycle and plays a key role in hydrological processes. Recently, satellite-based precipitation products (SPPs) have provided grid-based precipitation with spatiotemporal variability. However, SPPs contain a lot of uncertainty in estimated precipitation, and the spatial resolution of these products is still relatively coarse. To overcome these limitations, this study aims to generate new grid-based daily precipitation based on a combination of rainfall observation data with multiple SPPs for the period of 2003–2017 across South Korea. A Random Forest (RF) machine-learning algorithm model was applied for producing a new merged precipitation product. In addition, several statistical linear merging methods have been adopted to compare with the results achieved from the RF model. To investigate the efficiency of RF, rainfall data from 64 observed Automated Synoptic Observation System (ASOS) installations were collected to analyze the accuracy of products through several continuous as well as categorical indicators. The new precipitation values produced by the merging procedure generally not only report higher accuracy than a single satellite rainfall product but also indicate that RF is more effective than the statistical merging method. Thus, the achievements from this study point out that the RF model might be applied for merging multiple satellite precipitation products, especially in sparse region areas.

Malicious Traffic Flow Detection in IOT Using Ml Based Algorithms

Identifying the malicious traffic flows in Internet of things (IOT) is very important to monitor and avoid unwanted errors or the unwanted flows in the network. So, for a security to this network various machine learning algorithms (ML) has been introduced by various analyst to avoid this flow of error in the network. But, owing to the unsuitable selection of features, the ML models which introduced previously suffer from misclassify errors. So, there arises a need to study the problem of feature selection more depth to predict the accurate traffic flow observation in the network. To overcome this problem, a new structure in machine learning (ML) is introduced. So, for thisa novel features selection metric CorrAUC is suggested. So, based on this metric approach, a new feature selection algorithm CorrAUC is develop and design, it is based on wrapper technique to get features accurately by filtering to predict flow of traffic is suggested. Then, we applied multicriteria decision method called VIKOR which is used for validating the features selected for recognition the flow of traffic errors in the network. We estimate our approach by using the NSL-KDD dataset and three different ML algorithms.

Machine Learning Based Prediction of 1-year Survival after Isolated Heart Transplant

<h3>Purpose</h3> Given limitations of traditional scores of 1-year survival after isolated heart transplantation (HT), we sought to create machine learning algorithms (ML) and compare with traditional scores. <h3>Methods</h3> We included adults HT recipients from the UNOS database between 2010-2019. The study cohort was randomly split in a derivation and a validation dataset (3:1 ratio) that were used to train and test the following ML algorithms after feature selection. Additionally, a logistic regression model was implemented and the IMPACT score was calculated. Solely pre-transplant clinical and laboratory variables of the UNOS database were used for all risk scores. <h3>Results</h3> The study cohort comprised of 18,625 patients (53 ± 13 years, 73% males). At 1-year after cardiac transplant, there were 2334 (12.53%) deaths. A total of 134 pre-transplant variables were analyzed. Receiver-operator curves for all ML models and the IMPACT score are shown in Figure. Areas under the curve were 0.694, 0.650, 0.649, 0.647, 0.512 for the Adaboost, LR, DT, SVM, KNN models respectively, whereas the IMPACT score had an AUC of 0.569. <h3>Conclusion</h3> ML models created and validated by using a contemporary cohort of the UNOS database showed high accuracy in predicting 1-year survival after HT.

Application of Machine Learning Algorithms to Predict Central Lymph Node Metastasis in T1-T2, Non-invasive, and Clinically Node Negative Papillary Thyroid Carcinoma.

Purpose: While there are no clear indications of whether central lymph node dissection is necessary in patients with T1-T2, non-invasive, clinically uninvolved central neck lymph nodes papillary thyroid carcinoma (PTC), this study seeks to develop and validate models for predicting the risk of central lymph node metastasis (CLNM) in these patients based on machine learning algorithms.Methods: This is a retrospective study comprising 1,271 patients with T1-T2 stage, non-invasive, and clinically node negative (cN0) PTC who underwent surgery at the Department of Endocrine and Breast Surgery of The First Affiliated Hospital of Chongqing Medical University from February 1, 2016, to December 31, 2018. We applied six machine learning (ML) algorithms, including Logistic Regression (LR), Gradient Boosting Machine (GBM), Extreme Gradient Boosting (XGBoost), Random Forest (RF), Decision Tree (DT), and Neural Network (NNET), coupled with preoperative clinical characteristics and intraoperative information to develop prediction models for CLNM. Among all the samples, 70% were randomly selected to train the models while the remaining 30% were used for validation. Indices like the area under the receiver operating characteristic (AUROC), sensitivity, specificity, and accuracy were calculated to test the models' performance.Results: The results showed that ~51.3% (652 out of 1,271) of the patients had pN1 disease. In multivariate logistic regression analyses, gender, tumor size and location, multifocality, age, and Delphian lymph node status were all independent predictors of CLNM. In predicting CLNM, six ML algorithms posted AUROC of 0.70–0.75, with the extreme gradient boosting (XGBoost) model standing out, registering 0.75. Thus, we employed the best-performing ML algorithm model and uploaded the results to a self-made online risk calculator to estimate an individual's probability of CLNM (https://jin63.shinyapps.io/ML_CLNM/).Conclusions: With the incorporation of preoperative and intraoperative risk factors, ML algorithms can achieve acceptable prediction of CLNM with Xgboost model performing the best. Our online risk calculator based on ML algorithm may help determine the optimal extent of initial surgical treatment for patients with T1-T2 stage, non-invasive, and clinically node negative PTC.

Understanding the Effect of Hyperparameter Optimization on Machine Learning Models for Structure Design Problems

To relieve the computational cost of design evaluations using expensive finite element (FE) simulations, surrogate models have been widely applied in computer-aided engineering design. Machine learning algorithms (MLAs) have been implemented as surrogate models due to their capability of learning the complex interrelations between the design variables and the response from big datasets. Typically, an MLA regression model contains model parameters and hyperparameters. The model parameters are obtained by fitting the training data. Hyperparameters, which govern the model structures and the training processes, are assigned by users before training. There is a lack of systematic studies on the effect of hyperparameters on the accuracy and robustness of the surrogate model. In this work, we proposed to establish a hyperparameter optimization framework to deepen our understanding of the effect. Based on the sequential model-based optimization method, the Pareto front is generated by running the optimal acquisition and updating the surrogate model iteratively. The optimum acquisition works by repeating a design space shrinking process. Using the acquired optimum, the surrogate model is updated, which describes the relationship between the hyperparameter combinations (inputs) generated by Latin hypercube sampling from the design space and structural response (outputs) to evaluate the modeling accuracy. The updated model will then be used for the next iteration of optimal acquisition until the termination criterion is met. Four frequently used MLAs, namely Gaussian Process Regression (GPR), Support Vector Machine (SVM), Random Forest Regression (RFR), and Artificial Neural Network (ANN), are tested on four benchmark examples of structure design optimization. For each MLA model, the model accuracy and robustness before and after the hyperparameters optimization (HOpt) are compared. The results show that HOpt can generally improve the performance of the MLA models in general with dependency on model complexity. HOpt leads to unstable improvements in the MLAs accuracy and robustness for complex problems, which are featured by high-dimensional mixed-variable design space. We also investigated the additional computational costs incurred by HOpt. The training cost is closely related to the MLA architecture. After HOpt, the training cost of ANN and RFR is increased more than that of the GPR and SVM. In summary, this study benefits the selection of HOpt method for different types of design problems based on their complexity (i.e. design domain continuity and the number of design variables, etc.).

Identifying atmospheric pollutant sources using a machine learning dispersion model and Markov chain Monte Carlo methods

Estimating the sources of contaminant or hazard emissions is important for pollution control and safety management. Markov chain Monte Carlo (MCMC), combined with Bayesian inference, was used to identify the source terms of pollutants. However, the efficiency and accuracy of the forward dispersion model greatly impacted the performance of the estimation method. Therefore, a machine learning algorithm (MLA) model with high prediction accuracy and efficiency was proposed and coupled with MCMC method to estimate the source terms. A previously proposed MLA model was used to obtain the expected concentrations in Bayesian estimation. The Delayed Rejection Adaptive Metropolis (DRAM) method was applied to sample particles in order to form Markov chains. To evaluate the performance of the MCMC–MLA method, a Gaussian dispersion model was selected as the forward model. The performances of MCMC–MLA and MCMC–Gaussian models were then compared with release cases in Prairie Grass experiment and the results showed that the MCMC–MLA method converged more rapidly than the MCMC–Gaussian model. Nevertheless, release cases in the Round Hill experiment were also used to test the generalisability of the MCMC–MLA. The results indicated that the performance of MCMC–MLA was better than that of the MCMC–Gaussian model for estimating source terms in estimation accuracy. Hence, the MCMC–MLA method proposed here is potentially a useful tool for identifying emissions source parameters with high accuracy and efficiency, as well as reasonable probability estimates.

Corporate Default Forecasting with Machine Learning

We compare statistical models usually employed in credit risk forecasting with machine learning algorithms (ML). Using a large dataset which includes financial ratios and credit behavioral indicators for about 300,000 Italian non-financial firms from 2011 to 2017, we show that training the models on financial statement data only, ML models record a significant improvement in discriminatory power and precision with respect to statistical models; however, this improvement is less pronounced when we enlarge the training dataset to include also credit behavioral data

Abstract 16850: Machine Learning Based Estimation of Systolic and Diastolic Arterial Blood Pressure From the Electrocardiogram and Oxygen Saturation Waveforms

Introduction: Accurate, continuous blood pressure (BP) monitoring requires invasive equipment. We investigate the utility of a machine learning algorithm (ML) to estimate the systolic (sys) and diastolic (dia) BP from the electrocardiogram (ECG) and oxygen saturation (SpO2) signals. Methods: We identified over 2,000 hours (~9.2 M heart-beats), of 4 lead ECG, arterial BP, and SpO2 waveforms, from 82 random ICU patients, with various diagnoses. The ECG, BP and SpO2 waveforms were delineated using a wavelet-transform-based algorithm. Features extracted from the ECG and SpO2 signals were used as input to the ML model, while the sys/dia BP values of the corresponding BP signal (range: 0-250 mmHg), served as the gold standard. Using these features, we trained two independent Random Forest (RF) models to estimate the sys/dia BP values, and measured their performance over five-fold cross-validation. To identify the optimal window-length (# heart beats) in estimating sys/dia BP values, the entire process of extracting features and training/testing of RF models was performed for window-lengths of {1,5,6,7,8,9,10,11,12,13,14,15,20,30,50} heart-beats. Results: The root-mean-square error (RMSE) at each window-length is shown in the Figure. The lowest RMSE for both estimated sys/dia BP, 5.81±4.61 mmHg and 3.73±3.26 mmHg, respectively, was achieved at 14 heart-beats. The corresponding RMSE values using a one-beat length-window were 7.73±5.93 mmHg and 5.28±4.42 mmHg, respectively. This result indicates the ability to estimate the instantaneous BP values even on a beat-to-beat basis, with low-error. Conclusions: In comparison to prior studies, our novel ML model, to the best of our knowledge, has been trained over the largest amount of data. Our ML model can accurately estimate the sys/dia values of the arterial BP. Figure: RMSE values obtained while estimating the mean sys/dia BP, across different window-lengths.

P5723IEVA: Integration and Extraction of Variant Attributes in NGS analysis

Abstract Next Generation Sequencing (NGS) technology has taken a central role in the diagnosis of genetically-determined cardiovascular diseases. The differentiation of sequencing errors from real variants is a key-point of the genetic testing. In case of novel variants or variants of uncertain significance that may potentially impact on clinical decisions (e.g. ICD implantation in primary cardiomyopathies or preventive surgery in heritable aneurysmal diseases) it is crucial to exclude false positive (FP) and false negative (FN) sequence errors. To date, Sanger Sequencing is the gold standard tool used to confirm and validate NGS-identified variants. While FPs are excluded by the Sanger confirmation and the damage is for wasted costs and time, the FNs are non-resolvable errors because they are undetected and, obviously, not searched in Sanger confirmation, with the risk of missing genetic diagnoses. Purpose This project aimed at reducing NGS errors through the introduction of a bioinformatic solution in the bioinformatic analytic step of the genetic testing process. iEVA is a tool that enhances NGS-derived informative features to use them in a filtering process based on a Machine Learning algorithm (ML). It considers sequencing features (e.g. technical errors, duplicates of PCR) together with nucleotide sequence characteristics. Methods To demonstrate the effectiveness of iEVA in eliminating FP and FN errors from the NGS bioinformatic pipeline, we developed two ML-based filtering algorithms. The training dataset consisted of 7968 single nucleotide variants (SNV) and 306 Insertions and Deletions (InDels) validated by Sanger sequencing performed by expert molecular biologists. Variants derived from 800 sequences obtained with the Illumina Trusight Cardio panel containing 174 genes related to cardiovascular diseases. Two Random Forest classifiers were trained with the task of discriminating between sequencing error and real variant. The first one was trained using attributes derived from the most common variant caller (GATK v3.8), and the second one using iEVA results. To evaluate ML models, we used a 3-Fold cross-Validation and validated the results using an independent validation dataset consisting of 3415 SNV and 132 InDels. Results Using iEVA attributes, we obtained 1 FP (excluded by Sanger) and 3 FN (confirmed by Sanger) less than using common variant caller attributes. In the independent validation dataset, the iEVA-trained classification model identified 1 Sanger-confirmed variant that was missed by variant caller-trained model. Conclusions Variant filtering is crucial to exclude sequencing errors and to recognize true variants. Even a single filtering error may negatively impact on the patient when a genetic diagnosis is missed. To obtain a certain genetic diagnosis, a 0% error probability is needed. The introduction of iEVA in the pipeline is an easy, time- and cost-saving tool to reduce errors and to improve the precision of the genetic data. Acknowledgement/Funding Italian Ministry of Health Research Funding to the IRCCS Foundation University Hospital Policlinico San Matteo of Pavia

Remote Estimation of Mangrove Aboveground Carbon Stock at the Species Level Using a Low-Cost Unmanned Aerial Vehicle System

There is ongoing interest in developing remote sensing technology to map and monitor the spatial distribution and carbon stock of mangrove forests. Previous research has demonstrated that the relationship between remote sensing derived parameters and aboveground carbon (AGC) stock varies for different species types. However, the coarse spatial resolution of satellite images has restricted the estimated AGC accuracy, especially at the individual species level. Recently, the availability of unmanned aerial vehicles (UAVs) has provided an operationally efficient approach to map the distribution of species and accurately estimate AGC stock at a fine scale in mangrove areas. In this study, we estimated mangrove AGC in the core area of northern Shenzhen Bay, South China, using four kinds of variables, including species type, canopy height metrics, vegetation indices, and texture features, derived from a low-cost UAV system. Three machine-learning algorithm models, including Random Forest (RF), Support Vector Regression (SVR), and Artificial Neural Network (ANN), were compared in this study, where a 10-fold cross-validation was used to evaluate each model’s effectiveness. The results showed that a model that used all four type of variables, which were based on the RF algorithm, provided better AGC estimates (R2 = 0.81, relative RMSE (rRMSE) = 0.20, relative MAE (rMAE) = 0.14). The average predicted AGC from this model was 93.0 ± 24.3 Mg C ha−1, and the total estimated AGC was 7903.2 Mg for the mangrove forests. The species-based model had better performance than the considered canopy-height-based model for AGC estimation, and mangrove species was the most important variable among all the considered input variables; the mean height (Hmean) the second most important variable. Additionally, the RF algorithms showed better performance in terms of mangrove AGC estimation than the SVR and ANN algorithms. Overall, a low-cost UAV system with a digital camera has the potential to enable satisfactory predictions of AGC in areas of homogenous mangrove forests.