Bayes Point Machine Research Articles

Improving rule-based classifiers by Bayes point aggregation

The widespread adoption of artificial intelligence systems with continuously higher capabilities is causing ethical concerns. The lack of transparency, particularly for state-of-the-art models such as deep neural networks, hinders the applicability of such black-box methods in many domains, like the medical or the financial ones, where model transparency is a mandatory requirement, and hence white-box models are largely preferred over potentially more accurate but opaque techniques.For this reason, in this paper, we focus on ruleset learning, arguably the most interpretable class of learning techniques. Specifically, we propose Bayes Point Rule Classifier, an ensemble methodology inspired by the Bayes Point Machine, to improve the performance and robustness of rule-based classifiers. In addition, to improve interpretability, we propose a technique to retain the most relevant rules based on their importance, thus increasing the transparency of the ensemble, making it easier to understand its decision-making process.We also propose FIND-RS, a greedy ruleset learning algorithm that, under mild conditions, guarantees to learn hypothesis with perfect accuracy on the training set while preserving a good generalization capability to unseen data points.We performed extensive experimentation showing that FIND-RS achieves state-of-the-art classification performance at the cost of a slight increase in the ruleset complexity w.r.t. the competitors. However, when paired with the Bayes Point Rule Classifier, FIND-RS outperforms all the considered baselines.

Sentiment analysis of epidemiological surveillance reports on COVID-19 in Greece using machine learning models.

The present research deals with sentiment analysis performed with Microsoft Azure Machine Learning Studio to classify Facebook posts on the Greek National Public Health Organization (EODY) from November 2021 to January 2022 during the pandemic. Positive, negative and neutral sentiments were included after processing 300 reviews. This approach involved analyzing the words appearing in the comments and exploring the sentiments related to daily surveillance reports of COVID-19 published on the EODY Facebook page. Moreover, machine learning algorithms were implemented to predict the classification of sentiments. This research assesses the efficiency of a few popular machine learning models, which is one of the initial efforts in Greece in this domain. People have negative sentiments toward COVID surveillance reports. Words with the highest frequency of occurrence include government, vaccinated people, unvaccinated, telephone communication, health measures, virus, COVID-19 rapid/molecular tests, and of course, COVID-19. The experimental results disclose additionally that two classifiers, namely two class Neural Network and two class Bayes Point Machine, achieved high sentiment analysis accuracy and F1 score, particularly 87% and over 35%. A significant limitation of this study may be the need for more comparison with other research attempts that identified the sentiments of the EODY surveillance reports of COVID in Greece. Machine learning models can provide critical information combating public health hazards and enrich communication strategies and proactive actions in public health issues and opinion management during the COVID-19 pandemic.

An Ensemble Approach Based on Fuzzy Logic Using Machine Learning Classifiers for Android Malware Detection

In this study, a fuzzy logic-based dynamic ensemble (FL-BDE) model was proposed to detect malware exposed to the Android operating system. The FL-BDE model contains a structure that combines both the processing power of machine learning (ML)-based methods and the decision-making power of the Mamdani-type fuzzy inference system (FIS). In this structure, six different methods, namely, logistic regression (LR), Bayes point machine (BPM), boosted decision tree (BDT), neural network (NN), decision forest (DF) and support vector machine (SVM) were used as ML-based methods to benefit from their scores. However, through an approach involving the process of voting and routing, the scores of only three ML-based methods which were more successful in classifying either the negative instances or positive instances were sent to the FIS to be combined. During the combining process, the FIS processed the incoming inputs and determined the malicious application score. Experimental studies were performed by applying the FL-BDE model and ML-based methods to the balanced dataset obtained from the APK files downloaded in the Drebin database and Google Play Store. The obtained results showed us that the FL-BDE model had a much better performance than the ML-based models did, with an accuracy of 0.9933, a recall of 1.00, a specificity of 0.9867, a precision of 0.9868, and an F-measure of 0.9934. These results also proved that the proposed model can be used as a more competitive and powerful malware detection model compared to those of similar studies in the literature.

Development of a predictive inpatient falls risk model using machine learning.

The aims of this study were to create a model that detects the population at risk of falls taking into account a fall prevention variable and to know the effect on the model's performance when not considering it. Traditionally, instruments for detecting fall risk are based on risk factors, not mitigating factors. Machine learning, which allows working with a wider range of variables, could improve patient risk identification. The sample was composed of adult patients admitted to the Internal Medicine service (total, n = 22,515; training, n = 11,134; validation, n = 11,381). A retrospective cohort design was used and we applied machine learning technics. Variables were extracted from electronic medical records electronic medical records. The Two-Class Bayes Point Machine algorithm was selected. Model-A (with a fall prevention variable) obtained better results than Model-B (without it) in sensitivity (0.74 vs. 0.71), specificity (0.82 vs. 0.74), and AUC (0.82 vs. 0.78). Fall prevention was a key variable. The model that included it detected the risk of falls better than the model without it. We created a decision-making support tool that helps nurses to identify patients at risk of falling. When it is integrated in the electronic medical records, it decreases nurses' workloads by not having to collect information manually.

Modeling of Groundwater Potential Using Cloud Computing Platform: A Case Study from Nineveh Plain, Northern Iraq

Knowledge of the groundwater potential, especially in an arid region, can play a major role in planning the sustainable management of groundwater resources. In this study, nine machine learning (ML) algorithms—namely, Artificial Neural Network (ANN), Decision Jungle (DJ), Averaged Perceptron (AP), Bayes Point Machine (BPM), Decision Forest (DF), Locally-Deep Support Vector Machine (LD-SVM), Boosted Decision Tree (BDT), Logistic Regression (LG), and Support Vector Machine (SVM)—were run on the Microsoft Azure cloud computing platform to model the groundwater potential. We investigated the relationship between 512 operating boreholes with a specified specific capacity and 14 groundwater-influencing occurrence factors. The unconfined aquifer in the Nineveh plain, Mosul Governorate, northern Iraq, was used as a case study. The groundwater-influencing factors used included elevation, slope, curvature, topographic wetness index, stream power index, soil, land use/land cover (LULC), geology, drainage density, aquifer saturated thickness, aquifer hydraulic conductivity, aquifer specific yield, depth to groundwater, distance to faults, and fault density. Analysis of the contribution of these factors in groundwater potential using information gain ratio indicated that aquifer saturated thickness, rainfall, hydraulic conductivity, depth to groundwater, specific yield, and elevation were the most important factors (average merit > 0.1), followed by geology, fault density, drainage density, soil, LULC, and distance to faults (average merit < 0.1). The average merits for the remaining factors were zero, and thus, these factors were removed from the analysis. When the selected ML classifiers were used to estimate groundwater potential in the Azure cloud computing environment, the DJ and BDT models performed the best in terms of all statistical error measures used (accuracy, precision, recall, F-score, and area under the receiver operating characteristics curve), followed by DF and LD-SVM. The probability of groundwater potential from these algorithms was mapped and visualized into five groundwater potential zones: very low, low, moderate, high, and very high, which correspond to the northern (very low to low), southern (moderate), and middle (high to very high) portions of the study area. Using a cloud computing service provides an improved platform for quickly and cheaply running and testing different algorithms for predicting groundwater potential.

Comparative Analysis of Machine learning Methods to Identify signs of suspicious Transactions of Credit Institutions and Their Clients

In the field of financial monitoring, it is necessary to promptly obtain objective assessments of economic entities (in particular, credit institutions) for effective decision-making. Automation of the process of identifying unscrupulous credit institutions based on machine learning methods will allow regulatory authorities to quickly identify and suppress illegal activities. The aim of the research is to substantiate the possibilities of using machine learning methods and algorithms for the automatic identification of unscrupulous credit institutions. It is required to select a mathematical toolkit for analyzing data on credit institutions, which allows tracking the involvement of a bank in money laundering processes. The paper provides a comparative analysis of the results of processing data on the activities of credit institutions using classification methods — logistic regression, decision trees. The author applies support vector machine and neural network methods, Bayesian networks (Two-Class Bayes Point Machine), and anomaly search — an algorithm of a One-Class Support Vector Machine and a PCA-Based Anomaly Detection algorithm. The study presents the results of solving the problem of classifying credit institutions in terms of possible involvement in money laundering processes, the results of analyzing data on the activities of credit institutions by methods of detecting anomalies. A comparative analysis of the results obtained using various modern algorithms for the classification and search for anomalies is carried out. The author concluded that the PCA-Based Anomaly Detection algorithm showed more accurate results compared to the One-Class Support Vector Machine algorithm. Of the considered classification algorithms, the most accurate results were shown by the Two-Class Boosted Decision Tree (AdaBoost) algorithm. The research results can be used by the Bank of Russia and Rosfinmonitoring to automate the identification of unscrupulous credit institutions

Improved prediction of clay soil expansion using machine learning algorithms and meta-heuristic dichotomous ensemble classifiers

Soil swelling-related disaster is considered as one of the most devastating geo-hazards in modern history. Hence, proper determination of a soil’s ability to expand is very vital for achieving a secure and safe ground for infrastructures. Accordingly, this study has provided a novel and intelligent approach that enables an improved estimation of swelling by using kernelised machines (Bayesian linear regression (BLR) & bayes point machine (BPM) support vector machine (SVM) and deep-support vector machine (D-SVM)); (multiple linear regressor (REG), logistic regressor (LR) and artificial neural network (ANN)), tree-based algorithms such as decision forest (RDF) & boosted trees (BDT). Also, and for the first time, meta-heuristic classifiers incorporating the techniques of voting (VE) and stacking (SE) were utilised. Different independent scenarios of explanatory features’ combination that influence soil behaviour in swelling were investigated. Preliminary results indicated BLR as possessing the highest amount of deviation from the predictor variable (the actual swell-strain). REG and BLR performed slightly better than ANN while the meta-heuristic learners (VE and SE) produced the best overall performance (greatest R2 value of 0.94 and RMSE of 0.06% exhibited by VE). CEC, plasticity index and moisture content were the features considered to have the highest level of importance. Kernelized binary classifiers (SVM, D-SVM and BPM) gave better accuracy (average accuracy and recall rate of 0.93 and 0.60) compared to ANN, LR and RDF. Sensitivity-driven diagnostic test indicated that the meta-heuristic models’ best performance occurred when ML training was conducted using k-fold validation technique. Finally, it is recommended that the concepts developed herein be deployed during the preliminary phases of a geotechnical or geological site characterisation by using the best performing meta-heuristic models via their background coding resource.

Prediction of the mortality rate and framework for remote monitoring of pregnant women based on IoT

To save the risk during the pregnancies in remote areas (where women cannot approach the doctors in the urban areas for proper check-ups), the authors have proposed an IoT based remote monitoring of pregnant ladies where the data is collected at the cloud server. Machine learning techniques will be applied on the trimesters’ attributes to find out the reasons of mortality rate of the babies. The use of these advanced technologies in pregnant women care environment can absolutely eradicate the pregnancy complications and harmful incidents. An initial work towards this study is to assess mortality risk prediction in pregnant ladies using machine-learning algorithms for proper prediction and treatment on time. A dataset of 10,000 pregnant women is analysed in this study. Classification algorithms are used to check the death rate of new born babies based on the mother’s age. The survival ratio is presented by applying the various algorithms. Two class SVM model is presented as the most accurate prediction model which outperformed over boosted decision tree, average perceptron, decision forest, locally deep SVM, Bayes point machine, decision forest and logistic regression. Age of the mother and frequency of the pregnancy without proper gap (i.e. less than 3 years) make impact on the mortality rate of the babies. Among the various algorithms, the AUC value of decision forest is augmented at 91.4% whereas two class SVM shows significantly improved performance to 96.6%. The authors have proposed machine learning-based final model, death rate of infants, frequency of pregnancy and age of the mother were interrelated as notable risk factors for mortality in Indian ladies along with other issues.

WITHDRAWN: A benchmarking study of Parkinson's disease classification base on speech symptom features

In numerous prominent fields, such as e-business, communications, retail, and medical, information exploration in databases has built its success rate. The mining of medical data has immense potential to investigate the out-of-sight trends in the respective medical databases. This paper aims to provide a benchmarking analysis of the classification basis for Parkinson's disease (PD) on speech symptoms utilizing a data mining approach. The Mentation, Tasks of Everyday Life (ADL), Motor Analysis, and Therapy Symptoms will captivate various PD main signs. The speech symptom that is an ADL is a frequent cause for the disease's advancement. The dataset PD classification is accessed from UCI, an open archive of broad datasets of machine learning (ML). In this paper, a comparative review is carried out on various classification algorithms by adding the feature significance analysis and the accuracy analysis to the best classification algorithm. Subsequently, a collection of ML algorithms to classify PD are Averaged Perceptron (AP), Bayes Point Ma-chine (BPM), Boosted Decision Trees (BDT), Convolutional Neural Network (CNN), Decision Forest (DF), Logistic Regression (LR), Neural Network (NN), and Support Vector Machine (SVM). Each algorithm is tested and evaluated via a 5-fold cross-validation approach. The results show that the BDT achieves the highest accuracy of 90.03%, precision of 93.81%, recall of 93.32%, and AUC of 95.67. It outperforms all the other seven tested ML classifiers; nevertheless, the other classifiers also manage to produce high PD diagnosis results, which indicates the suitability of the speech analysis approach in PD diagnosis.

On the use of textual feature extraction techniques to support the automated detection of refactoring documentation

Refactoring is the art of improving the internal structure of a program without altering its external behavior, and it is an important task when it comes to software maintainability. While existing studies have focused on the detection of refactoring operations by mining software repositories, little was done to understand how developers document their refactoring activities. Therefore, there is recent trend trying to detect developers documentation of refactoring, by manually analyzing their internal and external software documentation. However, these techniques are limited by their manual process, which hinders their scalability. Hence, in this study, we tackle the detection of refactoring documentation as binary classification problem. We focus on the automatic detection of refactoring activities in commit messages by relying on text-mining, natural language preprocessing, and supervised machine learning techniques. We design our tool to overcome the limitation of the manual process, previously proposed by existing studies, through exploring the transformation of commit messages into features that are used to train various models. For our evaluation, we use and compare five different binary classification algorithms, and we test the effectiveness of these models using an existing dataset of manually curated messages that are known to be documenting refactoring activities in the source code. The experiments are carried out with different data sizes and number of bits. As per our results, the combination of Chi-Squared with Bayes point machine and Fisher score with Bayes point machine could be the most efficient when it comes to automatically identifying refactoring text patterns in commit messages, with an accuracy of 0.96, and an F-score of 0.96.

COMPARATIVE ANALYSIS OF CLASSIFICATION METHODS IN PROCESSING DATA ON THE ACTIVITIES OF CREDIT ORGANIZATIONS

The results of solving the classification problem of credit organizations from the point of view of possible involvement in the money laundering processes are presented. A comparative analysis of the results obtained using various modern classification algorithms is carried out. When analyzing credit institutions, Rosfinmonitoring analysts have to operate with large amounts of information. The actual need for the number of objects to be analyzed is in many times greater than the capabilities of analysts. This problematic situation requires prioritization of inspections. The heterogeneous nature of information resources and their significant volume exclude the possibility of their manual processing. It is necessary to move from successive expert examinations of individual objects to parallel mass automated checks, taking into account modern methodological and instrumental possibilities in the context of digital transformation of public administration. A comparative analysis of the results of processing data on the activities of credit organizations by classification methods – logistic regression, decision trees (algorithms of Two-Class Boosted Decision Forest, AdaBoost), the method of support vectors (algorithm of Two-Class Support Vector Machine), neural network methods (algorithm of Two-Class Neural Network), Bayesian networks (the algorithm of Two-Class Bayes Pointmachine) carried out. Of the classification algorithms considered, the most accurate results were shown by the algorithm of Two-Class Boosted Decision Forest (AdaBoost). The results obtained are of great practical importance and may allow Rosfinmonitoring analysts, as well as experts of the Bank of Russia, to identify deviant credit institutions potentially involved in money laundering processes.

Comparison of nomogram with machine learning techniques for prediction of overall survival in patients with tongue cancer

BackgroundThe prediction of overall survival in tongue cancer is important for planning of personalized care and patient counselling. ObjectivesThis study compares the performance of a nomogram with a machine learning model to predict overall survival in tongue cancer. The nomogram and machine learning model were built using a large data set from the Surveillance, Epidemiology, and End Results (SEER) program database. The comparison is necessary to provide the clinicians with a comprehensive, practical, and most accurate assistive system to predict overall survival of this patient population. MethodsThe data set used included the records of 7596 tongue cancer patients. The considered machine learning algorithms were logistic regression, support vector machine, Bayes point machine, boosted decision tree, decision forest, and decision jungle. These algorithms were mainly evaluated in terms of the areas under the receiver-operating characteristic (ROC) curve (AUC) and accuracy values. The performance of the algorithm that produced the best result was compared with a nomogram to predict overall survival in tongue cancer patients. ResultsThe boosted decision-tree algorithm outperformed other algorithms. When compared with a nomogram using external validation data, the boosted decision tree produced an accuracy of 88.7% while the nomogram showed an accuracy of 60.4%. In addition, it was found that age of patient, T stage, radiotherapy, and the surgical resection were the most prominent features with significant influence on the machine learning model’s performance to predict overall survival. ConclusionThe machine learning model provides more personalized and reliable prognostic information of tongue cancer than the nomogram. However, the level of transparency offered by the nomogram in estimating patients’ outcomes seems more confident and strengthened the principle of shared decision making between the patient and clinician. Therefore, a combination of a nomogram – machine learning (NomoML) predictive model may help to improve care, provides information to patients, and facilitates the clinicians in making tongue cancer management-related decisions.

A smart approach for fire prediction under uncertain conditions using machine learning

One of the most ubiquitous cause of worldwide deforestation and devastation of wildlife is fire. To control fire and reach the forest area in time is not always possible. Consequently, the level of destruction is often high. Therefore, predicting fires well in time and taking immediate action is of utmost importance. However, traditional fire prediction approaches often fail to detect fire in time. Therefore, a more reliable approach like the Internet of Things (IoT) needs to be adopted. IoT sensors can not only observe the real-time conditions of an area, but it can also predict fire when combined with Machine learning. This paper provides an insight into the use of Machine Learning models towards the occurrence of forest fires. In this context, eight Machine Learning algorithms: Boosted Decision Trees, Decision Forest Classifier, Decision Jungle Classifier, Averaged Perceptron, 2-Class Bayes Point Machine, Local Deep Support Vector Machine (SVM), Logistic Regression and Binary Neural Network model have been implemented. Results suggest that the Boosted decision tree model with the Area Under Curve (AUC) value of 0.78 is the most suitable candidate for a fire prediction model. Based on the results, we propose a novel IoT-based smart Fire prediction system that would consider both meteorological data and images for early fire prediction.

A Machine Learning Algorithm to Predict the Probability of (Occult) Posterior Malleolar Fractures Associated With Tibial Shaft Fractures to Guide "Malleolus First" Fixation.

To develop an accurate machine learning (ML) predictive model incorporating patient, fracture, and trauma characteristics to identify individual patients at risk of an (occult) PMF. Databases of 2 studies including patients with TSFs from 2 Level 1 trauma centers were combined for analysis. Using ten-fold cross-validation, 4 supervised ML algorithms were trained in recognizing patterns associated with PMFs: (1) Bayes point machine; (2) support vector machine; (3) neural network; and (4) boosted decision tree. Performance of each ML algorithm was evaluated and compared based on (1) C-statistic; (2) calibration slope and intercept; and (3) Brier score. The best-performing ML algorithm was incorporated into an online open-access prediction tool. Total data set included 263 patients, of which 28% had a PMF. Training of the Bayes point machine resulted in the best-performing prediction model reflected by good C-statistic, calibration slope, calibration intercept, and Brier score of 0.89, 1.02, -0.06, and 0.106, respectively. This prediction model was deployed as an open-access online prediction tool. A ML-based prediction model accurately predicted the probability of a (occult) PMF in patients with a TSF based on patient- and fracture-specific characteristics. This prediction model can guide surgeons in their diagnostic workup and preoperative planning. Further research is required to externally validate the model before implementation in clinical practice. Prognostic Level III. See Instructions for Authors for a complete description of levels of evidence.

Predicting fatalities among shark attacks: comparison of classifiers

<span lang="EN-US">This paper presents the comparisons of different classifiers on predicting Shark attack fatalities. In this study, we are comparing two classifiers which are Support vector machines(SVMs) and Bayes Point Machines(BPMs) on Shark attacks dataset. The comparison of the classifiers were based on the accuracy, recall, precision and F1-score as the performance measurement. The results obtained from this study showed that BPMs predicted the fatality of shack attack victim experiment with higher accuracy and precision than the SVMs because BPMs have “average” identifier which can minimize the probabilistic error measure. From this experiment, it is concluded that BPMs are more suitable in predicting fatality of shark attack victim as BPMs is an improvement of SVMs.</span>

Azure Machine Learning tools efficiency in the electroencephalographic signal P300 standard and target responses classification

Abstract The Event-Related Potentials were investigated on a group of 70 participants using the dense array electroencephalographic amplifier with photogrammetry geodesic station. The source localisation was computed for each participant. The activity of brodmann areas (BAs) involved in the brain cortical activity of each participant was measured. Then the mean electric charge flowing through particular areas was calculated. The five different machine learning tools (logistic regression, boosted decision tree, Bayes point machine, classic neural network and averaged perceptron classifier) from the Azure ecosystem were trained, and their accuracy was tested in the task of distinguishing standard and target responses in the experiment. The efficiency of each tool was compared, and it was found out that the best tool was logistic regression and the boosted decision tree in our task. Such an approach can be useful in eliminating somatosensory responses in experimental psychology or even in establishing new communication protocols with mildly mentally disabled subjects.

Development of a machine learning algorithm predicting discharge placement after surgery for spondylolisthesis.

We aimed to develop a machine learning algorithm that can accurately predict discharge placement in patients undergoing elective surgery for degenerative spondylolisthesis. The National Surgical Quality Improvement Program (NSQIP) database was used to select patients that underwent surgical treatment for degenerative spondylolisthesis between 2009 and 2016. Our primary outcome measure was non-home discharge which was defined as any discharge not to home for which we grouped together all non-home discharge destinations including rehabilitation facility, skilled nursing facility, and unskilled nursing facility. We used Akaike information criterion to select the most appropriate model based on the outcomes of the stepwise backward logistic regression. Four machine learning algorithms were developed to predict discharge placement and were assessed by discrimination, calibration, and overall performance. Nine thousand three hundred and thirty-eight patients were included. Median age was 63 (interquartile range [IQR] 54-71), and 63% (n = 5,887) were female. The non-home discharge rate was 18.6%. Our models included age, sex, diabetes, elective surgery, BMI, procedure, number of levels, ASA class, preoperative white blood cell count, and preoperative creatinine. The Bayes point machine was considered the best model based on discrimination (AUC = 0.753), calibration (slope = 1.111; intercept = - 0.002), and overall model performance (Brier score = 0.132). This study has shown that it is possible to create a predictive machine learning algorithm with both good accuracy and calibration to predict discharge placement. Using our methodology, this type of model can be developed for many other conditions and (elective) treatments. These slides can be retrieved under Electronic Supplementary Material.

Can Machine-learning Techniques Be Used for 5-year Survival Prediction of Patients With Chondrosarcoma?

Several studies have identified prognostic factors for patients with chondrosarcoma, but there are few studies investigating the accuracy of computationally intensive methods such as machine learning. Machine learning is a type of artificial intelligence that enables computers to learn from data. Studies using machine learning are potentially appealing, because of its possibility to explore complex patterns in data and to improve its models over time. The purposes of this study were (1) to develop machine-learning algorithms for the prediction of 5-year survival in patients with chondrosarcoma; and (2) to deploy the best algorithm as an accessible web-based app for clinical use. All patients with a microscopically confirmed diagnosis of conventional or dedifferentiated chondrosarcoma were extracted from the Surveillance, Epidemiology, and End Results (SEER) Registry from 2000 to 2010. SEER covers approximately 30% of the US population and consists of demographic, tumor characteristic, treatment, and outcome data. In total, 1554 patients met the inclusion criteria. Mean age at diagnosis was 52 years (SD 17), ranging from 7 to 102 years; 813 of the 1554 patients were men (55%); and mean tumor size was 8 cm (SD 6), ranging from 0.1 cm to 50 cm. Exact size was missing in 340 of 1544 patients (22%), grade in 88 of 1544 (6%), tumor extension in 41 of 1544 (3%), and race in 16 of 1544 (1%). Data for 1-, 3-, 5-, and 10-year overall survival were available for 1533 (99%), 1512 (98%), 1487 (96%), and 977 (63%) patients, respectively. One-year survival was 92%, 3-year survival was 82%, 5-year survival was 76%, and 10-year survival was 54%. Missing data were imputed using the nonparametric missForest method. Boosted decision tree, support vector machine, Bayes point machine, and neural network models were developed for 5-year survival. These models were chosen as a result of their capability of predicting two outcomes based on prior work on machine-learning models for binary classification. The models were assessed by discrimination, calibration, and overall performance. The c-statistic is a measure of discrimination. It ranges from 0.5 to 1.0 with 1.0 being perfect discrimination and 0.5 that the model is no better than chance at making a prediction. The Brier score measures the squared difference between the predicted probability and the actual outcome. A Brier score of 0 indicates perfect prediction, whereas a Brier score of 1 indicates the poorest prediction. The Brier scores of the models are compared with the null model, which is calculated by assigning each patient a probability equal to the prevalence of the outcome. Four models for 5-year survival were developed with c-statistics ranging from 0.846 to 0.868 and Brier scores ranging from 0.117 to 0.135 with a null model Brier score of 0.182. The Bayes point machine was incorporated into a freely available web-based application. This application can be accessed through https://sorg-apps.shinyapps.io/chondrosarcoma/. Although caution is warranted, because the prediction model has not been validated yet, healthcare providers could use the online prediction tool in daily practice when survival prediction of patients with chondrosarcoma is desired. Future studies should seek to validate the developed prediction model. Level III, prognostic study.

Analysing repeated hospital readmissions using data mining techniques

Few studies have examined how to identify future readmission of patients with a large number of repeat emergency department (ED) visits. We explore 30-day readmission risk prediction using Microsoft’s AZURE machine learning software and compare five classification methods: Logistic Regression, Boosted Decision Trees (BDTs), Support Vector Machine (SVM), Bayes Point Machine (BPM), and Two-Class Neural Network (TCNN). We predict the last readmission visit of frequent ED patients extracted from the electronic health records of their 8455 penultimate visits. The methods show differential improvement, with the BDT indicating marginally better AUC (area under the ROC curve) than logistic regression and BPM, followed by the TCNN and SVM. A comparison of BDT and Logistic Regression results for correct and incorrect classification highlights the similarities and differences in the significant predictors identified by each method. Future research may incorporate time-varying covariates to identify other longitudinal factors that can lead to readmission risk reduction.

Development of Machine Learning Algorithms for Prediction of 5-Year Spinal Chordoma Survival

Chordomas are locally invasive slow-growing tumors that are difficult to study because of the rarity of the tumors and the lack of significant volumes of patients with longitudinal follow-up. As such, there are currently no machine learning studies in the chordoma literature. The purpose of this study was to develop machine learning models for survival prediction and deploy them as open access web applications as a proof of concept for machine learning in rare nervous system lesions. The National Cancer Institute's Surveillance, Epidemiology, and End Results program database was used to identify adult patients diagnosed with spinal chordoma between 1995 and 2010. Four machine learning models were used to predict 5-year survival for spinal chordoma and assessed by discrimination, calibration, and overall performance. The 5-year overall survival for 265 patients with spinal chordoma was 67.5%. Variables used for prediction were age at diagnosis, tumor size, tumor location, extent of tumor invasion, and extent of surgery. For 5-year survival prediction, the Bayes Point Machine achieved the best performance with a c statistic of 0.80, calibration slope of 1.01, calibration intercept of 0.03, and Brier score of 0.16. This model for 5-year mortality prediction was incorporated into an open access application and can be found online (https://sorg-apps.shinyapps.io/chordoma/). This analysis of patients with spinal chordoma demonstrated that machine learning models can be developed for survival prediction in rare pathologies and have the potential to serve as the basis for creation of decision support tools in the future.