关键词： 大于胎龄儿   风险预测模型   机器学习   XGBoost  

摘要： 目的:尝试应用机器学习方法建立大于胎龄儿的风险预测模型。方法:回顾分析2013年1月1日至2014年12月31日于在浙江大学医学院附属妇产科医院规范产检并分娩的12683例孕妇的临床资料,包括孕妇年龄、身高、孕前体质量、孕期增重和各项化验结果及新生儿体重,运用Logistic回归和XGBoost两种机器学习算法构建大于胎龄儿的预测模型,评估这两种模型对大于胎龄儿预测的准确性。结果:XGBoost模型的受试者操作曲线的曲线下面积较大(0.92),Logistic回归模型较小(0.72),两者比较差异有统计学意义(P<0.05)。XGBoost模型的约登指数(0.63)、灵敏度(0.83)、特异度(0.80)均优于Logistic回归模型(0.34、0.65、0.68)(P均<0.05)。Logistic回归模型及XGBoost模型的阳性预测值均较低,分别为0.15、0.27。结论:机器学习方法(尤其是XGBoost)构建的模型预测大于胎龄儿的综合性能较强,具有潜在应用价值,值得进一步研究。

关键词： 脂肪肝   主成分分析   决策树   神经网络   支持向量机   贝叶斯网络   随机森林  

摘要： 目的:实现和比较5种常见的机器学习算法在脂肪肝分类预测研究中的应用。方法:通过主成分分析对数据的体检指标进行降维,然后应用决策树、神经网络、支持向量机、贝叶斯网络和随机森林算法,分别构建脂肪肝分类预测模型,对1956例体检数据进行脂肪肝的分类预测。结果:决策树分类模型在脂肪肝分类预测上的准确率最高,达到70.14%,其次是支持向量机和神经网络模型,处于68%左右的水平。结论:本文所研究的典型算法在脂肪肝分类的预测上具有较为可靠分类预测能力,但决策树模型在应用小样本数据上表现出了优势;同时,还发现臀围(HIP)和甘油三酯(TG)可能与脂肪肝分类关联密切。

关键词： 信用风险   刚性数据   机器学习方法   神经网络  

摘要： 大数据、人工智能技术的运用为企业信用风险预测提供了新的视角。本文通过爬虫程序抓取全市场约9000家企业的司法诉讼信息、招聘信息、法人和核心团队工商变更记录等无法粉饰的刚性数据,并采用机器学习等人工智能方法,开发出违约概率预测模型,可对企业违约概率做出前瞻性预测,从而实现对企业信用风险更早的预警。

关键词： 乳腺癌   Logistic回归   随机森林   敏感度  

摘要： 收集乳腺癌的预后状况,通过Logistic回归模型、决策树和随机森林算法,分析比较不同变量对乳腺癌预后的敏感性,并比较特征变量的重要性。使用Logistic回归的多变量分析显示,年龄、N-stage、乳腺癌内分泌治疗、化疗手段、多灶性病灶、化疗与预后有关。随机森林分析的结果得到了各个因素指标的重要性评分,有助于正确选择高风险人群,有利于建立分类系统。通过该结果找到危险因素,建立较高准确率的预测模型,预测乳腺癌患者的生存能力,并可作为制定医疗决策的参考。根据患者的特征制定针对患者不同特征的个体化术后随访及辅助治疗策略,争取最大程度上预防乳腺癌的术后复发。

关键词： 机器学习   消费金融   反欺诈   进展综述  

摘要： 消费金融机构向个人提供信贷额度或资金,具有客群下沉、 流程便捷等特点.相比其他传统金融业务,欺诈在消费金融行业更为严重,是妨碍行业发展的关键痛点.本文以构建欺诈侦测模型面临的五大关键挑战为线索,概述了机器学习方法在消费金融反欺诈模型上的应用进展.

关键词： 主成分分析法   过采样算法   贫困生   XGBoost模型   分类预测  

摘要： 针对高校贫困生认定工作中存在的问题,利用校园一卡通数据,综合学生消费和生活规律,结合XGBoost(Extreme Gradient Boosting)模型和主成分分析法、过采样算法,建立高校贫困生的分类预测方法。该方法在贫困生分类预测中的准确率较高。实验结果证明,采用XGBoost模型比其他模型预测准确率更高,为我国高校贫困学生的评定标准提供重要依据,保证了贫困学生认定工作的公正性。

关键词： Downward shortwave radiation   Machine learning methods   AVHRR   GBRT  

摘要： Downward shortwave radiation (DSR) is one of the major driving forces of climate system. Knowledge of the Earth's radiation budget is essential for improving our understanding of the Earth's climate. Therefore, accurate estimation of DSR has great significance. Satellite remote sensing is a practical way to derive DSR with high spatial resolution and coverage. In this study, four machine learning methods, including gradient boosting regression tree (GBRT), random forest (RF), multivariate adaptive regression spline (MARS), and artificial neural network (ANN), were applied to estimate DSR at a spatial resolution of 5 km and a temporal resolution of 1 day using Advanced Very High Resolution Radiometer (AVHRR) data. The DSR estimates based on four machine learning methods were evaluated using ground measurements at 96 sites over China. The measurements were collected from the Climate Data Center of the Chinese Meteorological Administration (CDC/CMA) from 2001 to 2003. The evaluation results showed that the GBRT method performed best at both daily and monthly time scales under both clear and cloudy sky conditions. The validation results at the daily time scale showed an overall root mean square error (RMSE) of 30.34 W m(-2) and an R value of 0.90 under clear sky conditions, whereas these values were 42.03 W m(-2) and 0.86, respectively, under cloudy sky conditions. The DSR estimates had an overall RMSE value of 16.93 W m(-2) and an R value of 0.97 at the monthly time scale. The Clouds and Earth's Radiant Energy System (CERES) Energy Balanced and Filled (EBAF) data sets were also used for comparison with the DSR estimates based on the GBRT method. The DSR estimates based on the GBRT method exhibited similar spatial distributions with those of the CERES-EBAF DSR product. Moreover, the DSR estimates based on the GBRT method did not show a clear overestimation tendency, as the CERES-EBAF DSR product did, at the CDC/CMA sites.

关键词： direct horizontal irradiance   physically-based model   machine learning method   estimation   China  

摘要： Accurate estimation of direct horizontal irradiance (DHI) is a prerequisite for the design and location of concentrated solar power thermal systems. Previous studies have shown that DHI observation stations are too sparsely distributed to meet requirements, as a result of the high construction and maintenance costs of observation platforms. Satellite retrieval and reanalysis have been widely used for estimating DHI, but their accuracy needs to be further improved. In addition, numerous modelling techniques have been used for this purpose worldwide. In this study, we apply five machine learning methods: back propagation neural networks (BP), general regression neural networks (GRNN), genetic algorithm (Genetic), M5 model tree (M5Tree), multivariate adaptive regression splines (MARS);and a physically based model, Yang's hybrid model (YHM). Daily meteorological variables, including air temperature (T), relative humidity (RH), surface pressure (SP), and sunshine duration (SD) were obtained from 839 China Meteorological Administration (CMA) stations in different climatic zones across China and were used as data inputs for the six models. DHI observations at 16 CMA radiation stations were used to validate their accuracy. The results indicate that the capability of M5Tree was superior to BP, GRNN, Genetic, MARS and YHM, with the lowest values of daily root mean square (RMSE) of 1.989 MJ m(-2)day(-1), and the highest correlation coefficient (R = 0.956), respectively. Then, monthly and annual mean DHI during 1960-2016 were calculated to reveal the spatiotemporal variation of DHI across China, using daily meteorological data based on the M5tree model. The results indicated a significantly decreasing trend with a rate of -0.019 MJ m(-2)during 1960-2016, and the monthly and annual DHI values of the Tibetan Plateau are the highest, while whereas the lowest values occur in the southeastern part of the Yunnan-Guizhou Plateau, the Sichuan Basin and most of the southern Yangtze Rive

关键词： Machine learning   Text mining   Biomedical effects of nanomaterials   Natural language processing   PubMed database  

摘要： Recently, the studies on biomedical effects of nanomaterials have already achieved many progresses tightly relating to human health. But the large amounts of research achievements in nanomaterial journals bomb too massive data to clarify the research highlight prediction. Fortunately, automatic text mining methods can complete the extracting information from a large set of documents efficiently by machining learning methods. We used both Naive Bayes and K-means clustering algorithms on manually labeled research data sets. It is 88.62% by the Naive Bayes algorithm classification result of 5-folds cross validation on sampled libraries. By applying the optimized Naive Bayes classification model, we made the research highlight trend prediction based on research achievements of biomedical effects of nanomaterials in 22 cutting edge nanomaterial journals including 350,000 original literatures in period from 2000 to 2017. The data mining clarified the polymer nanomaterial is the most researched nanomaterial but with a decreasing trend. The research interests of metallic and carbon based nanomaterial follow the polymer one, and possess increasing trend. We could predict that the research highlight trend on biomedical effects of nanomaterials is focused on polymer, metallic and carbon based material systems in the near future. (C) 2018 Published by Elsevier B.V.

关键词： machine learning   nursing informatics   Omaha System   public health nursing  

摘要： Background Public health nurses (PHNs) engage in home visiting services and documentation of care services for at-risk clients. To increase efficiency and decrease documentation burden, it would be useful for PHNs to identify critical data elements most associated with patient care priorities and outcomes. Machine learning techniques can aid in retrospective identification of critical data elements. Objective We used two different machine learning feature selection techniques of minimum redundancy-maximum relevance (mRMR) and LASSO (least absolute shrinkage and selection operator) and elastic net regularized generalized linear model (glmnet in R). Methods We demonstrated application of these techniques on the Omaha System database of 205 data elements (features) with a cohort of 756 family home visiting clients who received at least one visit from PHNs in a local Midwest public health agency. A dichotomous maternal risk index served as the outcome for feature selection. Application Using mRMR as a feature selection technique, out of 206 features, 50 features were selected with scores greater than zero, and generalized linear model applied on the 50 features achieved highest accuracy of 86.2% on a held-out test set. Using glmnet as a feature selection technique and obtaining feature importance, 63 features had importance scores greater than zero, and generalized linear model applied on them achieved the highest accuracy of 95.5% on a held-out test set. Discussion Feature selection techniques show promise toward reducing public health nursing documentation burden by identifying the most critical data elements needed to predict risk status. Further studies to refine the process of feature selection can aid in informing PHNs' focus on client-specific and targeted interventions in the delivery of care.