关键词： SEER数据库   乳腺癌   Logistic回归   决策树   预后因素  

摘要： 目的以SEER数据库中1990—2014年间的乳腺癌数据为研究对象,利用机器学习方法,分析乳腺癌的预后因素,辅助医师对患者的预后进行有效评判。方法根据临床医师的建议,筛选了12个字段作为模型输入字段,以术后5年生存状况作为模型输出字段。首先利用单因素统计分析方法初步筛选预后因素,再分别利用logistic回归和决策树两种机器学习分类算法进行建模分析,藉此寻找影响乳腺癌5年预后的因素。采用十折交叉法组织样本数据,并利用过抽样和欠抽样技术进行样本的平衡处理;以灵敏度、特异度及ROC下的AUC等参数作为模型的评价指标。结果在12个模型输入字段中,肿瘤分期、肿瘤分级、肿瘤尺寸、雌激素水平、年龄分组、孕激素水平等因素对于乳腺肿瘤预后具有较大影响;在此两种模型下,模型测试集上的灵敏度和特异度均介于74.2%~78.2%之间,AUC均处于0.838~0.850之间。结论利用Logistic回归和决策树算法构建乳腺癌患者的优化预后模型,可辅助医师判断患者预后情况及治疗效果。

关键词： 城投债   matlab   BP神经网络   内部评级  

摘要： 近年来,金融平台及其发行的城投债在经济增长和社会发展方面发挥了非常重要的作用。然而,市场上与城市债务有关的信用风险并不明显,而且外部评级机构在发行的城投债方面普遍虚高评级。在这种情况下,通过建立一个逻辑系统,分析城投债偿债风险,尽可能选择具有代表性的指标,通过BP神经网络等智能算法来实现城投债评级,以计算模型预测的准确性。归根结底,从结果来看,神经网络模型BP已超越了传统方法的界限,并通过模拟非线性模型来模拟实际数据,从而超越了对线性模型的依赖。

关键词： 安全帽   佩戴行为   机器学习   实时检测  

摘要： 施工现场作业人员是否佩戴安全帽主要依靠人工检查,存在监管效率低、时效性差等问题,为了实时自动监管作业人员是否佩戴安全帽,提出1种基于机器学习的安全帽佩戴行为检测方法。首先利用深度学习YOLOv3算法检测出现场视频中的施工人员脸部位置,根据安全帽与人脸的关系估算出安全帽潜在区域;然后对安全帽潜在区域图像进行增强处理,使用HOG(方向梯度直方图)提取样本的特征向量;再利用SVM(机器学习的支持向量机)分类器对脸部上方是否有安全帽进行判断,进而实现对施工人员安全帽佩戴行为的实时检测与预警。以某高铁站施工现场为例进行验证,研究数据表明在施工通道和塔吊作业区域,该方法可实时有效检测出工人未佩戴安全帽的行为,识别率达90%。

关键词： 地震波形记录   地震预警系统   机器学习方法   判别   应用   特征提取器   噪声影响   地震波形特征  

摘要： 由于天然或人为干扰产生的局部脉冲式噪声影响,使得地震预警系统出现误报。为减少此类问题,我们利用美国南加州和日本的30万条地震波形记录,训练了一个生成式对抗网络(GAN)用于识别P波初至特征。我们将GAN判别器作为自动特征提取器,并利用70万条地震事件和噪声波形记录训练了一个随机森林分类器。结果表明,该方法可以辨别99.2%的P波和98.4%的噪声信号。其优越的性能有望极大地减少因局部脉冲式噪声而造成的误触发数量。我们的研究表明,GAN判别器能获取简洁有效的地震波形特征,可广泛应用于地震学研究。小结 地震预警系统受到脉冲式噪声信号的干扰(非真实地震信号),有时偶尔会出现误触发的情况。这会造成不必要的经济损失和公众恐慌。本文利用机器学习工具判别波形是由地震产生的还是由局部噪声源产生的。我们通过利用美国南加州和日本大约70万条波形来训练算法。研究表明,利用机器学习方法可以识别99.2%的地震和98.4%的噪声。该方法可以减少大量的误报,能显著改善地震预警系统的鲁棒性。

关键词： Machine learning   Neural networks   Classification   Active targets   Time projection chamber  

摘要： We evaluate machine learning methods for event classification in the Active-Target Time Projection Chamber detector at the National Superconducting Cyclotron Laboratory (NSCL) at Michigan State University. Currently, events of interest are selected via cuts in the track fitting stage of the analysis workflow. An explicit classification step to single out the desired reaction product would result in more accurate physics results as well as a faster analysis process. We tested binary and multi-class classification methods on data produced by the Ar-46(p,p) experiment run at the NSCL in September 2015. We found that fine-tuning a pre-trained convolutional neural network produced the most successful classifier of proton scattering events in the experimental data, when trained on both experimental and simulated data. We present results from this investigation and conclude with recommendations for event classification in future experiments.

关键词： 深层神经网络   弹性计算   网络加速   地震周期   周期模型   机器学习方法   几何结构   流变结构  

摘要： 认识重复地震周期活动的影响,其最大的挑战之一是大尺度的黏弹性地震周期模型的计算成本。对于计算密集型黏弹性代码,须在数以千计的时间和位置点上进行评估。因此,研究趋向于采用几个固定的流变结构和模型几何结构,考查所预测的大震后短时间(小于10年)内给定深度处不同时间的形变。训练一个深层神经网络来学习黏弹性解在任意时间、位置和大范围流变结构计算效率的表示,可快速且可靠地完成计算,并具有较高时空分辨率。结果表明,该机器学习方法可以将黏弹性计算速度提高50000%以上。这种加速量级将使得在更大的模型参数范围内、在比以前更大的时空尺度上和数千个地震周期内,模拟几何结构复杂的断层成为可能。

关键词： 机器学习   小微企业   融资特征  

摘要： 构建小微企业信用融资特征分析模型对于开展小微企业信贷业务具有重要价值和意义,有益于改善融资双方信息不对称的困境,有助于增加资金提供方、政府部门对小微企业信用信息的了解和把握,提升小微企业风险管理的能力。企业征信系统收录了大量的小微企业信用信息,这对缓解银企信息不对称、促进小微企业融资具有重要的作用。本文从企业征信系统中筛选了33305家符合条件的小微企业,运用支持向量机(Support Vector Machine,SVM)模型,初步找到了几个可以帮助小微企业增信、拓宽融资渠道的特征向量。从而进一步帮助金融机构挖掘潜在客户,优化客户结构。

关键词： Building early design   Parametric design   Machine learning   Environmental impact   Prediction intervals  

摘要： Decisions made at early design stage have major impacts on buildings' life-cycle environmental performance. However, when only a few parameters are determined in early design stages, the detailed design decisions may still vary significantly. This may cause same early design to have quite different environmental impacts. Moreover, default settings for unknown detailed design parameters clearly cannot cover all possible variations in impact, and Monte Carlo analysis is sometimes not applicable as parameters' probability distributions are usually unknown. Thus, uncertainties about detailed design make it difficult for existing environmental assessment methods to support early design decisions. Thus, this study developed a quantitative method using parametric design technology and machine learning algorithms for assessing buildings' environmental performance in early decision stages, considering uncertainty associated with detailed design decisions. The parametric design technology creates design scenarios dataset, then associated environmental performances are assessed using environmental assessment databases and building performance simulations. Based on the generated samples, a machine learning algorithm integrating fuzzy C-means clustering and extreme learning machine extracts the case-specific knowledge regarding designed buildings' early design associated with environmental uncertainty. Proposed method is an alternative but more generally applicable method to previous approaches to assess building's environmental uncertainty in early design stages.

关键词： Machine-learning   Support vector machine   Deep neural network   Black spot identification  

摘要： With the rapid development of economy and urbanization, the number of urban motor vehicles keeps increasing. Urban travel is more convenient, but the traffic safety problems are increasingly prominent. Traffic accident data include not only time and place, but also people, roads, vehicles and the surrounding environment. Traffic accident black spot is the spatial location of traffic accident concentrated distribution. Most of the traditional traffic accident black spot identification only considers time and space factors, ignoring other factors. Based on the traffic accident data of Suzhou Industrial Park, this paper makes a fusion analysis of the multi-source influencing factors involved in traffic accident black spot. According to the structured association characteristics of urban traffic accident big data, a support vector machine method based on maximizing the classification interval is used to train the complex model and optimal learning of accident black spots in the study area. The accuracy of black spot identification is improved. At the same time, aiming at the rapid growth of traffic accident multi source data, a black point identification algorithm based on deep neural network is proposed. The deep neural network of relevant data category information is established to verify the model's ability to identify accident black spots. A feature-based black spot identification method based on depth neural network is proposed. Furthermore, a dynamic adaptive machine learning architecture is built.

关键词： Breast cancer   Biomarker   Saliva   Polyamines   Metabolomics   Alternative decision tree  

摘要： Purpose The aim of this study is to explore new salivary biomarkers to discriminate breast cancer patients from healthy controls. Methods Saliva samples were collected after 9 h fasting and were immediately stored at - 80 degrees C. Capillary electrophoresis and liquid chromatography with mass spectrometry were used to quantify hundreds of hydrophilic metabolites. Conventional statistical analyses and artificial intelligence-based methods were used to assess the discrimination abilities of the quantified metabolites. A multiple logistic regression (MLR) model and an alternative decision tree (ADTree)-based machine learning method were used. The generalization abilities of these mathematical models were validated in various computational tests, such as cross-validation and resampling methods. Results One hundred sixty-six unstimulated saliva samples were collected from 101 patients with invasive carcinoma of the breast (IC), 23 patients with ductal carcinoma in situ (DCIS), and 42 healthy controls (C). Of the 260 quantified metabolites, polyamines were significantly elevated in the saliva of patients with breast cancer. Spermine showed the highest area under the receiver operating characteristic curves [0.766;95% confidence interval (CI) 0.671-0.840, P < 0.0001] to discriminate IC from C. In addition to spermine, polyamines and their acetylated forms were elevated in IC only. Two hundred each of two-fold, five-fold, and ten-fold cross-validation using different random values were conducted and the MLR model had slightly better accuracy. The ADTree with an ensemble approach showed higher accuracy (0.912;95% CI 0.838-0.961, P < 0.0001). These prediction models also included spermine as a predictive factor. Conclusions These data indicated that combinations of salivary metabolomics with the ADTree-based machine learning methods show potential for non-invasive screening of breast cancer.