基于机器学习的妊娠期糖尿病智能预测系统设计与实现

为帮助有效识别患有妊娠期糖尿病的孕妇,提前干预治疗,降低其潜在风险,设计并开发了妊娠期糖尿病智能预测系统。首先,在对比10种机器学习模型的基础上对临床数据集进行数据归一化和特征筛选,减少模型计算复杂度和一些使模型不稳定的特征;其次,使用Stacking算法对10种不同的机器学习模型进行集成,分别搭建2个集成模型Stacking1和Stacking2,比较集成模型的性能;最后,基于PyQt5设计妊娠期糖尿病智能预测系统,预测孕妇患妊娠期糖尿病的风险,并给出建议。结果表明：使用10个机器学习模型以及2个集成模型对GDM进行预测,发现GBDT的预测结果高于其他机器学习模型,集成模型Stacking2将多个异质学习器相结合表现出较高的准确性和可靠性,且评价指标Accuracy、Precision、Recall、AUC分别为0.900 9、0.901 2、0.900 7、0.900 7,均高于同类模型。智能预测系统能有效预测妊娠期糖尿病的风险,能够及早发现易患病人群,并提供妊娠期糖尿病的科普知识,从而加强对易患病人群的健康管理,降低妊娠期糖尿病发生的风险。     

浅谈降低采购成本的有效途径——包装材料的回收及再利用

包装的回收再利用属于公司逆向物流的一部分,通过包装材料的回收及再利用能有效降低公司的采购成本。本文简单介绍包装产品的回收、检测、再利用措施决策等回收再利用三大步骤,并进一步探讨了包装产品在企业的回收再利用系统的建立和其带来的经济、社会和环境效益,最后又阐述了包装回收再利用系统的运作主体选择。     

探索中国的企业文化

企业文化是一个企业的管理文化,是经济发展的深层推动力。文章分析了中国企业文化的特色及其发展趋势,提出了在建设中国企业文化实践中要防止的诸多不良倾向。     

事件营销在汽车市场中的应用剖析

文章通过有选择性的事例归纳事件营销的特点,结合汽车业实际,提出了在汽车业开展事件营销的注意点,并最终提出了发展希望。     

全球化背景下我国经济运行的安全性

为了在经济全球化的大潮中立于不败之地,确保经济运行的安全性,我们应坚持科学发展观,不断提高驾驭经济运行的能力;实事求是,敢于解决经济运行中的突出矛盾;突出重点,不断构建经济运行的安全体系等。     

论地勘费基数的优化使用

地勘费基数优化使用是贯彻、落实国务院给予属地化地勘队伍优惠政策的一项重要内容,正确理解地勘费基数的特点是用好用足这项优惠政策的关键.文章认为,地勘费基数有四个最重要的特点应当引起地勘单位各级领导和相关管理人员的高度重视.通过对地勘费特点的归纳分析,提出了管理好、利用好地勘费的建议和措施.     

三维多箱模型预测大气环境质量

建立了16个风向下的三维多箱模式,该模式充分考虑了污染源的空间变化、气象条件和周围环境对污染扩散的影响,以及干沉积和化学转化等物理化学机理对SO2的去除作用,根据质量守恒定律,得到了箱体方程,最后将该模式用于石家庄市区SO2质量浓度的预测,并对其进行了验证,结果表明多箱模型是预测市区和经济开发区的大气质量最佳方法之一。     

多特征融合的视频镜头分割

针对传统镜头分割算法特征单一、分割效果较差等问题，提出了一种多特征融合的视频镜头分割算法。首先，通过建立一种优化模型，融合空间差异度量和感知哈希度量两种重要特征，构造一种镜头边界区分能力更强的镜头边界检测特征，即像素差异度量；其次，结合直方图差异度量，形成有效的镜头边界检测策略，以便更加准确判断是否发生镜头切换。实验结果表明，与传统算法相比，所提算法取得了较好的分割效果，具有较高的查准率和查全率。     

Can users embed their user experience in user-generated images? Evidence from JD.com

Nowadays, massive user-generated images (UGIs) are posted online to convey users' experiences with specific brands or products. Thus, this visual information is precious, as it conveys users' actual and subjective feelings about brands and products. Because of the unprecedented quantity of images and the heterogeneity of their content, it is quite challenging for brand marketers and retailers to probe into subjective user experience in large-scale UGIs. To address this gap, this study aims to identify the connection between user experience and different image semantic features (i.e. centrality and richness) by using deep learning models. By employing objective data (8963 images) from JD.com and using deep learning algorithms (faster R-CNN), we found that users with positive user experience prefer to generate high-centrality and high-richness pictures. Our study enriches the relevant literature and provides valuable practical implications for brand marketers and e-commerce retailers. Based on findings of this work, relevant stakeholders can understand their users’ experience better from objective UGIs and devise corresponding recommendation and service strategies.     

A novel hotel recommendation method based on personalized preferences and implicit relationships

On tourism websites, hotel recommendations have drawn growing attention from researchers, as they can help customers select a satisfactory hotel from many options with massive information. However, some inherent challenges exist in conventional hotel recommendations, specifically the extent to which there is considerable room for improvement in user preference models and neighbour recognition. Therefore, we propose a two-stage hotel recommendation approach that employs hotel feature information to support preference analysis. First, in the filling stage, association rules between features are considered to accurately capture users’ personalized preferences, which can be incorporated with public preferences to estimate potential ratings of users for unvisited hotels. Then, in the recommendation stage, we combine rating similarities between users with their closeness relationships to identify more reliable neighbours. Finally, a hotel recommendation case on Ctrip.com is performed to evaluate the model. Experimental results confirm that our method outperforms the other five benchmark methods.