关键词： 信用卡违约预警模型   机器学习   过滤式特征选择   决策树   随机森林  

摘要： 文章使用机器学习中的决策树和随机森林算法建立信用卡违约预警模型。第一步,建立相关系数热力图,发现用户的违约状态与个人贷款的相关性最强、其次是开户时长和存款时长;第二步,使用过滤式特征选择方法选取主要的特征;第三步,分别使用决策树和随机模型建立信用卡违约预警模型;第四步,总结这两个模型的优缺点。结论:决策树模型训练速度快,但是模型精度一般;随机森林模型训练速度慢,但是模型精度优良。

关键词： 共享单车需求分析   岭回归   随机森林  

摘要： 共享单车是当前一种流行的出行方式,不同地区和不同时段对共享单车的需求不同。更好的共享单车需求分析能够优化共享单车配置,降低闲置率,提高利用率。本文依据Kaggle华盛顿共享单车使用的历史数据,对未来几天的共享单车需求进行预测,要求利用每月前19天的历史数据预测20天之后的共享单车租赁情况。这一问题属于机器学习中的回归问题,在Kaggle提供的数据集上分别采用了线性模型(Ridge Regression),基于回归树的集成学习模型(Random Forest)以及深度学习模型(Neural Network)。以预测结果和真实值的RMSE(Root Mean Squared Logarithmic Error)作为模型表现的评价指标。对不同的模型特点以及在共享单车数据集上的表现进行了对比分析,对结果进行了总结。

关键词： 岩浆岩   定量分类模型   成矿预测   大数据   北山  

摘要： 岩浆岩分类是研究岩浆岩成因及其成矿作用的基础,传统的岩浆岩二维图解分类,其结果不具唯一性。本文基于地质大数据研究方法;应用多元统计"聚类分析"和"判别分析"的算法,选取北山地区岩浆岩258件岩浆岩样本构建起"北山地区岩浆岩氧化物+微量元素+稀土元素定量分类系列模型",在35个变量空间将岩浆岩分为3大类共11个小类,与岩浆岩有关的金、铜、铅锌、铀钍、钨钼矿等岩体分别集中处于不同的分类中,充分证明所建定量分类系列模型对北山地区岩浆岩分类的高度有效性,同时也对该区岩浆岩找矿靶区做出了准确的定量预测。

关键词： 结构健康监测   超声导波   机器学习   随机森林回归   损伤监测  

摘要： 随着复合材料结构健康监测技术的广泛应用,基于先进传感与数据挖掘技术的结构损伤识别已成为国内外的研究热点问题。本文以机器学习理论技术为基础,对复合材料结构受拉伸疲劳载荷造成的损伤进行定量化识别分析和研究,针对结构疲劳损伤监测数据之间存在的时序相关性,构建出基于随机森林的结构损伤识别模型,为复合材料结构健康监测分析提供了新的思路和方法。主要研究内容如下:提取不同路径的信号特征信息,包括信号能量,信号峰值,信号主峰到达时间。选择MSE作为不纯度函数,设计并构建了一种基于随机森林的结构损伤识别模型;为了验证本文所提出的基于机器学习的复合材料结构损伤识别模型的有效性,以Bookshelf框架结构数据为基准进行对比试验分析。结果表明:基于信号能量,信号峰值,信号主峰到达时间特征的随机森林结构损伤识别模型可以有效识别与预测损伤。本文试验数据来自NASA与斯坦福大学进行的联合损伤表征试验。

关键词： Neurological disorders   Machine Learning   Computer-aided diagnosis   A Python toolbox for network classification   Autism spectrum disorder  

摘要： Background: Autism spectrum disorder (ASD) affects the brain connectivity at different levels. Nonetheless, non-invasively distinguishing such effects using magnetic resonance imaging (MRI) remains very challenging to machine learning diagnostic frameworks due to ASD heterogeneity. So far, existing network neuroscience works mainly focused on functional (derived from functional MRI) and structural (derived from diffusion MRI) brain connectivity, which might not directly capture relational morphological changes between brain regions. Indeed, machine learning (ML) studies for ASD diagnosis using morphological brain networks derived from conventional T1-weighted MRI are very scarce. New method: To fill this gap, we leverage crowdsourcing by organizing a Kaggle competition to build a pool of machine learning pipelines for neurological disorder diagnosis with application to ASD diagnosis using cortical morphological networks derived from T1-weighted MRI. Results: During the competition, participants were provided with a training dataset and only allowed to check their performance on a public test data. The final evaluation was performed on both public and hidden test datasets based on accuracy, sensitivity, and specificity metrics. Teams were ranked using each performance metric separately and the final ranking was determined based on the mean of all rankings. The first-ranked team achieved 70% accuracy, 72.5% sensitivity, and 67.5% specificity, where the second-ranked team achieved 63.8%, 62.5%, 65% respectively. Conclusion: Leveraging participants to design ML diagnostic methods within a competitive machine learning setting has allowed the exploration and benchmarking of wide spectrum of ML methods for ASD diagnosis using cortical morphological networks.

关键词： 氢原子钟   故障诊断   机器学习   嵌入式  

摘要： 氢原子钟作为一种精密的授时守时仪器,在科学研究与工程应用中发挥着重要作用,目前,我国原子钟还存在设备故障、可靠性差等问题。为了简化技术人员排查流程,提升维修效率,提出了一种采用机器学习算法,将氢钟历史运行数据作为训练样本,结合DBSCAN及人工神经网络算法得到氢钟诊断模型,从而大幅简化故障排查过程的方法。实验中将训练好的模型部署到嵌入式设备上,并把实时预判结果给技术人员作为参考,证实了这种方法的可行性及有效性。

关键词： Machine learning   Deflection calculation   BOTDR   Fiber optic sensing cable  

摘要： Due to its unique advantages, the distributed fiber optical sensing (DFOS) technology has been used to study the performance of inclinometer so as to monitor landslide deformation. Strain distribution of inclinometer can be obtained by distributed strain sensing (DSS) cables, and the strain-deflection relationship can be established by using the widely accepted methods (e.g., the quadratic integral method and classical conjugate beam method). However, the application of quadratic integral method and classical conjugate beam method are based on many assumptions, and there will be remarkable deviation between calculated deflection and actual displacement with the increase of integral length. Given this, a new deflection calculation method based on machine learning is proposed. Through learning on the monitoring data, an implicit function model between depth, strain, and measured displacement is established by using the BP (back propagation) neural network algorithm. The efficiency of the proposed model has been verified against measured displacement, which demonstrates the capability of this method for landslide deformation prediction. Compared with the traditional integral method, the lateral deflection curve of inclinometer calculated by the proposed method is closer to the actual measured displacement both in trend and values. The proposed model shows great potential in the application of deflection calculation in engineering.

关键词： Himawari-8 AOD   PM2   5 concentration   Machine learning method   Model combination  

摘要： The estimation of the PM2.5 concentration based on satellite remote sensing is currently a hot topic for research, with machine learning algorithms starting to be applied to the estimation model. However, there are few comparisons between different machine learning algorithms, and few scholars objectively evaluate the estimation performance of machine learning algorithms under different weather conditions. In this study, PM2.5 concentration was estimated by six commonly used machine learning algorithms, and their performances were compared in four different weather conditions in the Beijing-Tianjin-Hebei (BTH) region. The results showed that decision tree and random forest consistently performed well in different weather conditions, while SVM performed poorly. When the PM2.5 concentration was greater than 150 mu g/m(3), the R and RMSE values for decision tree were 0.854 and 31.53 mu g/m(3), respectively, while the evaluation coefficient of SVM algorithm was only 0.597 and 49.31 mu g/m(3), it was worth noting that all algorithms performed better in this interval than in others. This study also focused on the development of an optimal combination algorithm to estimate PM2.5 concentration under different weather conditions and got a good application effect. The results of this research may provide a theoretical basis and an important reference for the application of machine learning algorithms in the field of remote sensing.

关键词： Machine learning   MBA admission selection   Non-native language   Performance prediction  

摘要： Although the research on student selection criteria has been very rich up to now, the role of the level of foreign language played in the admission selection of a non-native spoken program is still receiving little attention. This study intends to explore the issue through three research methods: (1) two-sample test of a hypothesis;(2) multiple linear regression analysis;(3) machine learning algorithms (Ridge regression, SVM, Random forest, GBDT). The case about 549 students enrolled in the Shanghai International MBA Program in China from 2007 to 2014 was used as empirical research samples. Through three methods of analysis and comparison, it was found that Oral English fluency played a key role in the admission selection of the English spoken MBA program in China. It is confirmed that the criteria, such as Rank of the graduated university, Company Nature, Latest Highest Degree, Math Exam, Sponsor (Tuition provider) and Stress management, have very good effect in predicting the final grades of students when graduation. This study also shows that the methods based on machine learning algorithm modeling such as ridge regression and SVM are suitable for student selection decision modeling.

关键词： Radiation Oncology   DICOM   Machine Learning   Class imbalance   Random Forest   TG-263  

摘要： Purpose: To present a Machine Learning pipeline for automatically relabeling anatomical structure sets in the Digital Imaging and Communications in Medicine (DICOM) format to a standard nomenclature that will enable data abstraction for research and quality improvement. Methods: DICOM structure sets from approximately 1200 lung and prostate cancer patients across 40 treatment centers were used to build predictive models to automate the relabeling of clinically specified structure labels to standardized labels as defined by the American Association of Physics in Medicine's (AAPM) Task Group 263 (TG-263). Volumetric bitmaps were created based on the delineated volumes and were combined with associated bony anatomy data to build feature vectors. Feature reduction was performed with singular value decomposition and the resulting vectors were used for predicting the label of each structure using five different classifier algorithms on the Apache Spark platform with 5-fold cross-validation. Undersampling methods were used to deal with underlying class imbalance that hindered the performance of classifiers. Experiments were performed on both a curated version of the data, which included only annotated structures, and the non-curated data that included all structures from the original treatment plans. Results: Random Forest provided the highest accuracies with F, scores of 98.77 for lung and 95.06 for prostate on the curated data sets. Scores were lower with 95.67 for lung and 90.22 for prostate on the non-curated data sets, highlighting some of the challenges of classifying real clinical data. Including bony anatomy data and pooling information from all structures for the same patient both increased accuracies. In some cases, undersampling with k-Means clustering for class balancing improved classifier accuracy but in all experiments it significantly reduced run time compared to random undersampling. Conclusion: This work shows that structure sets can be relabeled using ou