Random Forest Importance Research Articles

Finding the combination of multiple biomarkers to diagnose oral squamous cell carcinoma – A data mining approach

Data mining has proven to be a reliable method to analyze and discover useful knowledge about various diseases, including cancer research. In particular, data mining and machine learning algorithms to study oral squamous cell carcinoma (OSCC), the most common form of oral cancer, is a new area of research. This malignant neoplasm can be studied using saliva samples. Saliva is an important biofluid that must be used to verify potential biomarkers associated with oral cancer. In this study, first, we provide an overview of OSSC diagnoses based on machine learning and salivary metabolites. To our knowledge, this is the first study to apply advanced data mining techniques to diagnose OSCC. Then, we give new results of classification and feature selection algorithms used to identify potential salivary biomarkers of OSCC. To accomplish this task, we used the filter feature selection random forest importance algorithm and a wrapper methodology to evaluate the importance of metabolites obtained from gas chromatography mass-spectrometry (GC-MS) in the context of differentiation of OSCC and the control group. Salivary samples (n = 68) were collected for the control group, and the OSCC group were from patients matched for gender, age, and smoking habit. The classification process occurred based on Random Forest (RF) classification algorithm along with 10-cross validation. The results showed that glucuronic acid, maleic acid, and batyl alcohol can classify the samples with an area under the curve (AUC) of 0.91 versus an AUC of 0.76 using all 51 metabolites analyzed. The methodology used in this study can assist healthcare professionals and be adopted to discover diagnostic biomarkers for other diseases.

Prognostic value of machine learning for acute heart failure

Abstract Funding Acknowledgements Type of funding sources: None. Background At onset of acute heart failure (AHF), various clinical fundamental parameters including vital sign, laboratory data, or initial treatment were investigated, and we can roughly estimate the prognosis. However, machine learning method for prediction of the prognosis was not studied. Purpose To elucidate prognostic value of machine learning for AHF comparing conventional statistical model. Methods We enrolled consecutive 300 patients with AHF (79.5 ± 12.1 years, 158 Males). Patients with acute coronary syndrome, mechanical circulatory support cases, and cardio-pulmonary arrest cases were excluded. The patients were randomly divided into 80% (240 cases) and 20% (60 cases), and the former was used as train data, and the latter as validation data. Objective variable was set as cardiac death in one year. First, logistic regression analysis with Akaike’s information criterion (AIC) was performed, and extracted predictive parameters. The predictive model for the cardiac prognosis was constructed by cut-off value of ROC curve analysis of propensity score was calculated. Next, machine learning (random forest method and deep learning) to build predictive model was performed with the predictors. Finally, accuracy of each predictive model was compared. Results Thirty cases showed cardiac death in one year. Logistic regression with AIC extracted 8 predictors, and the cut off-value of propensity score with the 6 parameters was 0.110. The accuracy was 0.714 and area under ROC (AUROC) was 0.836. Conversely, random forest method demonstrated the accuracy as 0.927, AUROC 0.860. On deep learning, the accuracy was 0.937 and AUROC 0.901. The top 4 high feature importance of random forest were Cl/red blood cell count/pH/Anion Gap. However, accuracy of those predictors was lower than that of machine learning. Conclusion Machine learning was a powerful tool to predict cardiac prognosis of AHF, comparing with conventional statistical model. Abstract Figure. Statistical model

Towards a software defect proneness model: feature selection

This article is focused on improving static models of software reliability based on using machine learning methods to select the software code metrics that most strongly affect its reliability.The study used a merged dataset from the PROMISE Software Engineering repository, which contained data on testing software modules of fiveprograms and twenty-onecode metrics. For the prepared sampling, the most important features that affect the quality of software code have been selected using the following methods of feature selection: Boruta, Stepwiseselection, Exhaustive Feature Selection, Random Forest Importance, LightGBM Importance, Genetic Algorithms, Principal Component Analysis, Xverse python.Basing on the voting on the results of the work of the methods of feature selection, a static (deterministic) model of software reliability has been built, which establishes the relationship between the probability of a defect in the software module and the metrics of its code. It has been shown that this model includes such code metrics as branch countof a program, McCabe’s lines of codeand cyclomatic complexity, Halstead’s total number of operators and operands, intelligence, volume, andeffort value.A comparison of the effectiveness of different methods of feature selection has been put into practice, in particular, a study of the effectof the method of feature selection on the accuracy of classification using the following classifiers: Random Forest, Support Vector Machine, k-Nearest Neighbors, Decision Tree classifier, AdaBoost classifier, Gradient Boosting for classification. It has been shown that the use of any method of feature selection increases the accuracy of classification by at least ten percentcompared to the original dataset, which confirms the importance of this procedure for predicting software defects based on metric datasets that contain a significant number of highly correlated software code metrics.It has been found that the best accuracy of the forecast for most classifiers was reachedusing a set of features obtained from the proposed static model of software reliability. In addition, it has been shown that it is also possible to use separate methods, such as Autoencoder,Exhaustive Feature Selection and Principal Component Analysis withan insignificant loss of classification and prediction accuracy.

Multidimensional sentiment recognition of film and television scene images

Identifying and analyzing the sentiment information of an image is a complicated task in content understanding with nondeterministic features. To establish an effective model for multidimensional emotion recognition of film and television scene images and explore the relationship between the inputs and prediction results, we chose eight feature selection schemes based on principal component analysis, least absolute shrinkage operator (Lasso) regression, Pearson correlation, and the importance of random forest (RF) to establish the input vectors and the relevant model of machining learning. After the test on the data set with fixed combination of dimension reduction of 19 emotional descriptors, the fuse of Lasso regression and the RF importance extraction were proved to have the best performance on accuracy with the regression models. The experimental results showed that our algorithm performed well on the regression task of multidimensional affective dataset, which was convenient and time-saving. The extracted features and relevant dimension reduction methods also achieved good results on the cross-dataset classification tasks.

Abstract 9495: Prognostic Value of Machine Learning on Clinical Parameters for Cardiac Prognosis in Patinets with Acute Congestive Heart Failure

Introduction: Congestive heart failure (CHF) in acute phase, many clinical fundamental parameters including vital sign, laboratory data, or initial treatment were investigated, and we can roughly estimate the prognosis with the parameters. However, machine learning method for prediction of the prognosis is still challenging. Hypothesis: Machine learning methods have higher performance for predicting cardiac prognosis in patients with CHF than conventional statistical model. Methods: We enrolled consecutive 300 patients with CHF (79.5±12.1 years, 158 Males). Patients with acute coronary syndrome, mechanical circulatory support cases, and cardio-pulmonary arrest cases were excluded. Objective variable was set as cardiac death in one year. First, multivariate stepwise logistic regression analysis with Akaike’s information criterion (AIC) was performed, and extracted parameters which had the most versatile predictive value. Propensity score (PS) was calculated with the extracted parameters, and the prediction model constructed by cut-off value of ROC curve analysis of PS. Next, machine learning (random forest method and deep learning) to build predictive model was performed with the predictors. Finally, accuracy of each predictive model was compared. Results: Thirty cases showed cardiac death in one year. Logistic regression with AIC extracted 8 predictors, and the cut off-value of propensity score with the 8 parameters was 0.110. The accuracy was 0.714 and area under ROC (AUROC) was 0.836. Conversely, random forest method demonstrated the accuracy as 0.927, AUROC 0.860. On deep learning, the accuracy was 0.937 and AUROC 0.901.The top 4 high feature importance of random forest were Cl/red blood cell count/pH/Anion Gap. However, accuracy of those predictors was lower than that of machine learning. Conclusions: Machine learning was a powerful tool to predict cardiac prognosis of CHF, comparing with conventional statistical model.

Abstract 9727: Predictive Value of Machine Learning on Adenosine Stress Single Photon Emission Computed Tomography for Multivessel Coronary Artery Stenosis

Introduction: Adenosine stress single photon emission computed tomography (SPECT) is utilized to detect coronary artery stenosis, but the diagnostic performance is recognized not to be enough in patients with multivessel disease (MVD). Estimation of left ventricular (LV) dysfunction on electrocardiography gated SPECT including LV dissynchrony was reported to be valuable to predict MVD, but prognostic value of machine learning of those data was not studied. Hypothesis: Machine learning of gated SPECT data has high diagnostic performance for MVD. Methods: We enrolled consecutive 256 patients (78±11 years, 188 male) who underwent adenosine stress gated SPECT (99mTechnesium) and coronary angiography in one month. LV function and myocardial perfusion score were evaluated by Heart Risk View (Nihon Medi-Physics Co.,Ltd.). First, multivariate logistic regression analysis underwent with minimum Akaike’s information critetion by stepwise regression. Propensity score (PS) was calculated with the selected predictors, and receiver operation characteristics (ROC) curve analysis was performed to evaluate the predictive value. Next, we constructed prediction model for MVD by machine learning (random forest method and deep learning). Results: Thirty-five cases (14%) showed MVD. Stepwise regression with AIC showed 7 predictors as the Figure. Their accuracy and area under ROC were in 0.527-0.809, 0.599-0.689, respectively. PS of the 7 predictors showed the accuracy 0.691 and AUROC 0.758. Random forest method and deep learning built prediction models, and their accuracy and AUROC were 0.840 and 0.788, 0.846 and 0.760, respectively. The top two feature importance of random forest were SD od TES on stress and Phase SD on stress, which indicate the degree of LV dyssynchrony. Conclusions: Machine learning of adenosine stress gated STECT had powerful predictive value for MVD, and the degree of LV dyssynchrony on stress played essential role in the prediction.

Feature Importance of Acute Rejection among Black Kidney Transplant Recipients by Utilizing Random Forest Analysis: An Analysis of the UNOS Database.

Background: Black kidney transplant recipients have worse allograft outcomes compared to White recipients. The feature importance and feature interaction network analysis framework of machine learning random forest (RF) analysis may provide an understanding of RF structures to design strategies to prevent acute rejection among Black recipients. Methods: We conducted tree-based RF feature importance of Black kidney transplant recipients in United States from 2015 to 2019 in the UNOS database using the number of nodes, accuracy decrease, gini decrease, times_a_root, p value, and mean minimal depth. Feature interaction analysis was also performed to evaluate the most frequent occurrences in the RF classification run between correlated and uncorrelated pairs. Results: A total of 22,687 Black kidney transplant recipients were eligible for analysis. Of these, 1330 (6%) had acute rejection within 1 year after kidney transplant. Important variables in the RF models for acute rejection among Black kidney transplant recipients included recipient age, ESKD etiology, PRA, cold ischemia time, donor age, HLA DR mismatch, BMI, serum albumin, degree of HLA mismatch, education level, and dialysis duration. The three most frequent interactions consisted of two numerical variables, including recipient age:donor age, recipient age:serum albumin, and recipient age:BMI, respectively. Conclusions: The application of tree-based RF feature importance and feature interaction network analysis framework identified recipient age, ESKD etiology, PRA, cold ischemia time, donor age, HLA DR mismatch, BMI, serum albumin, degree of HLA mismatch, education level, and dialysis duration as important variables in the RF models for acute rejection among Black kidney transplant recipients in the United States.

Research on Influencing Factors and Classification of Patients With Mild and Severe COVID-19 Symptoms.

ObjectiveTo analyze the epidemiological history, clinical symptoms, laboratory testing parameters of patients with mild and severe COVID-19 infection, and provide a reference for timely judgment of changes in the patients’ conditions and the formulation of epidemic prevention and control strategies.MethodsA retrospective study was conducted in this research, a total of 90 patients with COVID-19 infection who received treatment from January 21 to March 31, 2020 in the Ninth People’s Hospital of Dongguan City were selected as study subject. We analyzed the clinical characteristics of laboratory-confirmed patients with COVID-19, used the oversampling method (SMOTE) to solve the imbalance of categories, and established Lasso-logistic regression and random forest models.ResultsAmong the 90 confirmed COVID-19 cases, 79 were mild and 11 were severe. The average age of the patients was 36.1 years old, including 49 males and 41 females. The average age of severe patients is significantly older than that of mild patients (53.2 years old vs 33.7 years old). The average time from illness onset to hospital admission was 4.1 days and the average actual hospital stay was 18.7 days, both of these time actors were longer for severe patients than for mild patients. Forty-eight of the 90 patients (53.3%) had family cluster infections, which was similar among mild and severe patients. Comorbidities of underlying diseases were more common in severe patients, including hypertension, diabetes and other diseases. The most common symptom was cough [45 (50%)], followed by fever [43 (47.8%)], headache [7 (7.8%)], vomiting [3 (3.3%)], diarrhea [3 (3.3%)], and dyspnea [1 (1.1%)]. The laboratory findings of patients also included leukopenia [13(14.4%)] and lymphopenia (17.8%). Severe patients had a low level of creatine kinase (median 40.9) and a high level of D-dimer. The median NLR of severe patients was 2.82, which was higher than that of mild patients. Logistic regression showed that age, phosphocreatine kinase, procalcitonin, the lymphocyte count of the patient on admission, cough, fatigue, and pharynx dryness were independent predictors of COVID-19 severity. The classification of random forest was predicted and the importance of each variable was displayed. The variable importance of random forest indicates that age, D-dimer, NLR (neutrophil to lymphocyte ratio) and other top-ranked variables are risk factors.ConclusionThe clinical symptoms of COVID-19 patients are non-specific and complicated. Age and the time from onset to admission are important factors that determine the severity of the patient’s condition. Patients with mild illness should be closely monitored to identify those who may become severe. Variables such as age and creatine phosphate kinase selected by logistic regression can be used as important indicators to assess the disease severity of COVID-19 patients. The importance of variables in the random forest further complements the variable feature information.

Inferring and analyzing gene regulatory networks from multi-factorial expression data: a complete and interactive suite

BackgroundHigh-throughput transcriptomic datasets are often examined to discover new actors and regulators of a biological response. To this end, graphical interfaces have been developed and allow a broad range of users to conduct standard analyses from RNA-seq data, even with little programming experience. Although existing solutions usually provide adequate procedures for normalization, exploration or differential expression, more advanced features, such as gene clustering or regulatory network inference, often miss or do not reflect current state of the art methodologies.ResultsWe developed here a user interface called DIANE (Dashboard for the Inference and Analysis of Networks from Expression data) designed to harness the potential of multi-factorial expression datasets from any organisms through a precise set of methods. DIANE interactive workflow provides normalization, dimensionality reduction, differential expression and ontology enrichment. Gene clustering can be performed and explored via configurable Mixture Models, and Random Forests are used to infer gene regulatory networks. DIANE also includes a novel procedure to assess the statistical significance of regulator-target influence measures based on permutations for Random Forest importance metrics. All along the pipeline, session reports and results can be downloaded to ensure clear and reproducible analyses.ConclusionsWe demonstrate the value and the benefits of DIANE using a recently published data set describing the transcriptional response of Arabidopsis thaliana under the combination of temperature, drought and salinity perturbations. We show that DIANE can intuitively carry out informative exploration and statistical procedures with RNA-Seq data, perform model based gene expression profiles clustering and go further into gene network reconstruction, providing relevant candidate genes or signalling pathways to explore. DIANE is available as a web service (https://diane.bpmp.inrae.fr), or can be installed and locally launched as a complete R package.

Landslide susceptibility analyses using Random Forest, C4.5, and C5.0 with balanced and unbalanced datasets

The effects of landslides have been exponentially increasing due to the rapid growth of urbanization and global climate change. The information gained from predictive models and landslide susceptibility analyses can be used to develop warning systems and mitigation measures. A comparative study was conducted to evaluate the effectiveness of landslide susceptibility analyses in a given area using three decision tree algorithms including Random Forest (RF), C4.5, and C5.0. Two sets of imagery datasets (raster and vector) were used and three combinations of 13 conditioning factors (including seven geotechnical properties of the soil) were determined by Information Gain, Gain Ratio, Chi-Squared Test, and Random Forest Importance. Datasets for the landslide conditioning factors were created based on the outcomes from the feature selection methods, in three different scenarios. In Scenario 1 the least important factors/features (as identified by information gain, chi-square, and gain ratio measures) were eliminated. In Scenario 2 only the most important factors (as identified by RF feature selection method evaluation) were kept. In Scenario 3, no factor was eliminated, using the data directly obtained from the sources without applying any feature selection method. The performances of the models were evaluated using statistical verification scores. C4.5 was found to have the highest performance when all 13 conditioning parameters (Scenario 3) were used for both the raster and vector data set. The RF model was the least effective in predicting the landslides in all three scenarios. However, the use of the balance vector dataset significantly increased the performance of the RF model. C4.5 and C5.0 had significantly better performance in handling extremely unbalance data in comparison to RF. Density, silt and clay content, and Atterberg’s limits (LL and PI) were the most important geotechnical conditioning factors in the performed landslide susceptibility analyses.

The Optimization Model for Reducing RON Loss in Gasoline Refining Process

As gasoline is the main fuel of small vehicles, the exhaust emissions from its combustion will affect air quality. The focus of gasoline cleaning is to reduce the sulfur and olefin content in gasoline while maintaining its RON as much as possible. The reduction of RON will bring great economic losses to enterprises. Therefore, it is very important for petrochemical enterprises to construct a RON loss model in the gasoline refining process. The model construction, which reduces RON loss during gasoline refining, is the main question in this paper. By Python and SPSS software, we got two variable filtering methods: the random forest importance filtering and PCA filtering, and combined with SVR and random forest models, RON of the product and sulfur content were predicted. The filtering order of the original data by Excel and Python is maximum and minimum removal, 3σ criterion removal, deletion of too many sites in incomplete data, and filling of empty values in the mean within two hours. Several RON prediction models were established with the help of Python software, and the variables selected were compared by two filtering methods: one is the SVR model based on Gaussian, linear, polynomial, and Sigmoid kernel functions; the other is the random forest model. The sulfur content and RON prediction model was constructed, which use evaluation functions such as MSE, R 2 , and RMSE to evaluate and sulfur content as the subject condition. We convert the problem into linear and nonlinear model variable optimization problems: the linear model is the variable selected by the SVR linear kernel function model and random forest; the nonlinear model is the combination of variables selected by the random forest model and random forest. Optimizing for each sample, the optimization method is to find the optimal solution for each variable and use the optimal method for each variable as the local optimal solution for the sample. The two models are evaluated from the perspectives of optimization degree, optimization rate, model running speed, etc.

Ensemble method based architecture using random forest importance to predict employee’s turn over

The departure of a skilled employee can create a problem for a company and this incident is increasing globally. Employee turnover has become an important issue these days due to the heavy workload, low pay, low job satisfaction, poor working environment. Companies face problems as their budget will increase, losing skilled manpower and employees’ trust. It’s taking time to adjust for a new employee and bring risk and increase the cost for the company. It is necessary to bring appropriate solutions to the problem. The main purpose of this paper is to predict the turnover of employees with the help of state of the art machine learning classifier. We have determined employee turnover selection factors using some prediction models. We first pre-processed the dataset by removing correlative attributes. Then, we have scaled the attributes. Secondly, a Sequential selection algorithm (SBS) has been using to reduce features from a high number to a relatively small signal-canton. Then use Chi-square and Random Forest important algorithms to determine the most significant shared key features. Then we get average_montly_hours, satisfaction_level, time_spend_company are responsible for the employee’s departure. Then, we have applied different state of the art machine learning algorithm to measure the accuracy. We have achieved the highest accuracy of 99.4% using the reduced feature with 10-Fold Cross-validation by applied the Random Forest classifier and which is higher than the mentioned reference work.

Inferring mechanisms of response prioritization on social media under information overload

Human decision-making is subject to the biological limits of cognition. The fluidity of information propagation over online social media often leads users to experience information overload. This in turn affects which information received by users are processed and gain a response to, imposing constraints on volumes of, and participation in, information cascades. In this study, we investigate properties contributing to the visibility of online social media notifications by highly active users experiencing information overload via cross-platform social influence. We analyze simulations of a coupled agent-based model of information overload and the multi-action cascade model of conversation with evolutionary model discovery. Evolutionary model discovery automates mechanistic inference on agent-based models by enabling random forest importance analysis on genetically programmed agent-based model rules. The mechanisms of information overload have shown to contribute to a multitude of global properties of online information cascades. We investigate nine characteristics of online messages that may contribute to the prioritization of messages for response. Our results indicate that recency had the largest contribution to message visibility, with individuals prioritizing more recent notifications. Global popularity of the conversation originator had the second highest contribution, and reduced message visibility. Messages that presented opportunity for novel user interaction, yet high reciprocity showed to have relatively moderate contribution to message visibility. Finally, insights from the evolutionary model discovery results helped inform response prioritization rules, which improved the robustness and accuracy of the model of information overload.

Effect of Genetic Crossing and Nutritional Management on the Mineral Composition of Carcass, Blood, Leather, and Viscera of Sheep.

The effect of genetic crossing and nutritional management on weight gain and the concentration of minerals and trace elements in the carcass, blood, leather, and viscera of sheep were evaluated. Several statistical strategies were used to evaluate the different elemental composition characteristics of pure breed animals, i.e., White Dorper (ODO), Ile de France (OIF), Texel (OTX), and Santa Inês (OSI), as well as their crossbreeds 1/2 White Dorper and 1/2 Santa Inês (ODS), 1/2 Ile de France, and 1/2 Santa Inês (OIS), 1/2 Texel × 1/2 Santa Inês (OTS). Three different diets were evaluated AL (ad libitum), R75, and R63 (75 and 63g of dry matter/kg of the animal metabolic weight, respectively). Levels of Ca, Cu, Fe, K, Mg, Mn, P, S, and Zn were determined by inductively coupled plasma optical emission spectrometry (ICP OES). The concentration of inorganic elements in the different body components was not affected by the diet (P> 0.05), and OTX and OTS were the breeds with the highest weight gain. Random forest importance models demonstrated that Zn in the carcass, K, Ca, and Zn in blood, and K in leather are most associated with separating the different evaluated sheep's breeds. Crossbreeding the native Santa Inês breed with sheep of exotic breeds produces animals well adapted to confinement.

Predictive modeling for wine authenticity using a machine learning approach

The purpose of this paper is to classify wines from 4 different countries in South America. Each class of wines is formed by samples considered by experts as representatives of the following commercial categories: “Argentinean Malbec (AM)”, “Brazilian Merlot (BM)”, “Uruguayan Tannat (UT)” and “Chilean Carménère (CC)”. The 83 samples collected were analyzed according to their composition of volatiles, semi-volatiles and phenolic compounds. We built a decision system for classification based on support vector machines (SVM), along with Correlation-based Feature selection (CFS), and Random Forest Importance (RFI), which measures the relative importance of the input variables. First, we use CFS to select a subset of variables among 190 chemical compounds. Thirteen chemicals were selected as correlated to the category and uncorrelated with each other. Afterwards, these chemical compounds were organized according to the importance ranking given by the RFI and classified with SVM. The study clearly indicated that SVM in combination with feature selection methods was able to identify the most important chemicals to classify the wine samples. Among the compounds identified in the wine samples, the variable subset defined by the feature selection methods, which were catechin, gallic, octanoic acid, myricetin, caffeic, isobutanol, resveratrol, kaempferol, and ORAC, were able to achieve an accuracy of 93.97% in classifying the commercial categories.

Landslide Susceptibility Mapping Using Ant Colony Optimization Strategy and Deep Belief Network in Jiuzhaigou Region

Landslidesusceptibility mapping (LSM) is the primary link of geological disaster risk evaluation, which is significant for postearthquake emergency response and rebuilding after disasters. Existing LSM studies applying deep learning (DL) methods have shortcomings such as easy overfitting, slow convergence, and insufficient hyperparametric optimization. In response to these problems, this study proposes an ensemble model based on ant colony optimization strategy and deep belief network (ACO-DBN). In ACO-DBN, DL optimization strategies were added to DBN and their combined parameters were optimized with ACO. Next, Pearson's correlation coefficient and random forest importance ranking methods were utilized to optimize landslide causative factors. Then, the Jiuzhaigou earthquake region was selected as an example to explore the applicability of this model. Besides, we conducted the Wilcoxon signed rank test in order to verify that the differences were statistically significant. In a comprehensive comparative all indexes and landslide density, the model proposed in this article shows good rationality, scientificity, and interpretability. The newly occurred landslide site further demonstrates that heuristically optimized DL could make scientific and accurate evaluation of landslide susceptibility.

Body weight, serum albumin and food intolerance were linked to upper gastrointestinal Crohn's disease: a 7-year retrospective analysis.

BackgroundThe clinical features of upper gastrointestinal (L4) Crohn’s disease (CD) and its subtypes, along with the associated and nutritional status, remain poorly described. Our aim was to evaluate the clinical characteristics of L4 CD phenotype and its subtypes at diagnosis, and their relationship with the nutritional status.MethodsA retrospective study was conducted on 869 CD patients diagnosed between 2013 and 2019, and the association between the clinical characteristics and nutritional status of L4 patients was determined using Random forest importance ranking and logistic regression.ResultsThe majority of the patients (59.72%) presented L4 lesions, of which 335, 158 and 26 had proximal ileal, jejunal and esophago-gastroduodenal (EGD) lesions respectively. L4 patients were predominantly male (OR 2.07), smoker (OR 1.80), and had higher body weight and BMI, longer disease course, and stricturing disease (OR 1.88). Furthermore, the serum albumin level, body weight and disease course showed higher MDG in the random forest importance ranking test for L4 CD and L4-proximal ileal types. According to logistic regression, body weight (OR 1.054), disease course (OR 1.010), stricturing behavior (OR 4.998) and tomato intolerance (OR 1.313) were the independent risk factors for L4. In addition, body weight (OR 1.042) and stricturing behavior (OR 3.152) were the relevant factors for proximal ileal subtype, and stricturing behavior (OR 4.206) and perianal disease (OR 0.339) for jejunal subtype.ConclusionsL4 disease has a higher incidence rate compared to the non-L4 CD, and mainly affects males, and those with prolonged disease course, stricturing behavior, higher weight, BMI, albumin levels and food intolerance (FI).

Artificial Intelligence in Ovarian Cancer Diagnosis.

This study aimed to use artificial intelligence (AI) to predict the pathological diagnosis of ovarian tumors using patient information and data from preoperative examinations. A total of 202 patients with ovarian tumors were enrolled, including 53 with ovarian cancer, 23 with borderline malignant tumors, and 126 with benign ovarian tumors. Using 5 machine learning classifiers, including support vector machine, random forest, naive Bayes, logistic regression, and XGBoost, we derived diagnostic results from 16 features, commonly available from blood tests, patient background, and imaging tests. We also analyzed the importance of 16 features on the prediction of disease. The highest accuracy was 0.80 in the machine learning algorithm of XGBoost. The evaluation of importance of the features showed different results among the correlation coefficient of the features, the regression coefficient, and the features importance of random forest. AI could play a role in the prediction of pathological diagnosis of ovarian cancer from preoperative examinations.

Machine Learning Detects Pattern of Differences in Functional Magnetic Resonance Imaging (fMRI) Data between Chronic Fatigue Syndrome (CFS) and Gulf War Illness (GWI).

Background: Gulf War Illness (GWI) and Chronic Fatigue Syndrome (CFS) are two debilitating disorders that share similar symptoms of chronic pain, fatigue, and exertional exhaustion after exercise. Many physicians continue to believe that both are psychosomatic disorders and to date no underlying etiology has been discovered. As such, uncovering objective biomarkers is important to lend credibility to criteria for diagnosis and to help differentiate the two disorders. Methods: We assessed cognitive differences in 80 subjects with GWI and 38 with CFS by comparing corresponding fMRI scans during 2-back working memory tasks before and after exercise to model brain activation during normal activity and after exertional exhaustion, respectively. Voxels were grouped by the count of total activity into the Automated Anatomical Labeling (AAL) atlas and used in an “ensemble” series of machine learning algorithms to assess if a multi-regional pattern of differences in the fMRI scans could be detected. Results: A K-Nearest Neighbor (70%/81%), Linear Support Vector Machine (SVM) (70%/77%), Decision Tree (82%/82%), Random Forest (77%/78%), AdaBoost (69%/81%), Naïve Bayes (74%/78%), Quadratic Discriminant Analysis (QDA) (73%/75%), Logistic Regression model (82%/82%), and Neural Net (76%/77%) were able to differentiate CFS from GWI before and after exercise with an average of 75% accuracy in predictions across all models before exercise and 79% after exercise. An iterative feature selection and removal process based on Recursive Feature Elimination (RFE) and Random Forest importance selected 30 regions before exercise and 33 regions after exercise that differentiated CFS from GWI across all models, and produced the ultimate best accuracies of 82% before exercise and 82% after exercise by Logistic Regression or Decision Tree by a single model, and 100% before and after exercise when selected by any six or more models. Differential activation on both days included the right anterior insula, left putamen, and bilateral orbital frontal, ventrolateral prefrontal cortex, superior, inferior, and precuneus (medial) parietal, and lateral temporal regions. Day 2 had the cerebellum, left supplementary motor area and bilateral pre- and post-central gyri. Changes between days included the right Rolandic operculum switching to the left on Day 2, and the bilateral midcingulum switching to the left anterior cingulum. Conclusion: We concluded that CFS and GWI are significantly differentiable using a pattern of fMRI activity based on an ensemble machine learning model.

Unbiased variable importance for random forests

The default variable-importance measure in random forests, Gini importance, has been shown to suffer from the bias of the underlying Gini-gain splitting criterion. While the alternative permutation importance is generally accepted as a reliable measure of variable importance, it is also computationally demanding and suffers from other shortcomings. We propose a simple solution to the misleading/untrustworthy Gini importance which can be viewed as an over-fitting problem: we compute the loss reduction on the out-of-bag instead of the in-bag training samples.