关键词： 癌症分类   癌症基因组图谱(TCGA)   深度森林  

摘要： 乳腺癌是癌症中较为常见的一种,拥有很高的死亡率。乳腺癌早期的诊断对于治疗有着至关重要的意义,现代医学对于乳腺癌的通常使用医学影像,病理分析等方法进行初期的诊断。随着新一代的测序技术的发展,基因与转录组数据的获得越来越容易。基因转录组数据结合机器学习算法的运用,可以快速、准确地检测出癌症患病风险。提出一种基于基因转录组的特征选择与分类预测的算法,所提出的算法在The Cancer Genome Atlas(TCGA)中的乳腺癌数据集中进行验证,实验结果能够精确地预测分类信息。

关键词： Boruta算法   随机森林   支持向量机   k最近邻   梯度提升   回归问题  

摘要： 近年来,机器学习方法在各个领域内的应用十分广泛,而特征挖掘则是机器学习过程中十分重要的数据预处理过程。Boruta算法是一种基于随机森林算法的特征筛选算法,该算法可以对机器学习过程中的自变量进行筛选,并且进行重要性排序。随机森林、支持向量机、k最近邻算法和梯度提升模型则是机器学习中的经典算法,在解决回归和分类问题中均有较广泛的应用。本文通过红葡萄酒案例,将Boruta算法与4种机器学习方法相结合,探讨Boruta算法的作用及几种不同机器学习方法的区别,为将特征挖掘算法和机器学习在回归问题中的应用提供参考。

关键词： Decision trees  

摘要： It is significant to integrate geochemical and geological survey to search deep gold orebody in modern gold mineralization prospecting. However, geological survey is laborious and expensive. Therefore, it is beneficial to study the results of the previous geological survey and find the regular pattern in the massive and complex gold deposit data. In this research, Logistic regression, Random forest and Decision tree are applied to train the model using raw data and resampling data. Recall, precision and accuracy are also used to evaluate model performance. The confusion matrix is selected to visualize the performance of different models. After comparison, it is found that the gold mineralization data cannot be identified because of the huge imbalance between the two groups of data. The test data are going to be predicted into the group with huge data. Random Forest has a good performance on the resampling data, recall and accuracy are 90.63% and 70.78%, respectively. The influence of the different classification boundaries is also discussed. According to the different requirements, the values of the different classification boundaries can be chosen. Evaluating with different measures may improve the adequacy of the model for gold mineralization prediction and improve the efficiency of the gold survey. © 2020, Science Press. All right reserved.

关键词： Sentinel-2   Sentinel-1   ALOS-2 PALSAR-2   mangrove   above-ground biomass   extreme gradient boosting regression   genetic algorithm   North Vietnam  

摘要： This study proposes a hybrid intelligence approach based on an extreme gradient boosting regression and genetic algorithm, namely, the XGBR-GA model, incorporating Sentinel-2, Sentinel-1, and ALOS-2 PALSAR-2 data to estimate the mangrove above-ground biomass (AGB), including small and shrub mangrove patches in the Red River Delta biosphere reserve across the northern coast of Vietnam. We used the novel extreme gradient boosting decision tree (XGBR) technique together with genetic algorithm (GA) optimization for feature selection to construct and verify a mangrove AGB model using data from a field survey of 105 sampling plots conducted in November and December of 2018 and incorporated the dual polarimetric (HH and HV) data of the ALOS-2 PALSAR-2 L-band and the Sentinel-2 multispectral data combined with Sentinel-1 (C-band VV and VH) data. We employed the root-mean-square error (RMSE) and coefficient of determination (R-2) to evaluate the performance of the proposed model. The capability of the XGBR-GA model was assessed via a comparison with other machine-learning (ML) techniques, i.e., the CatBoost regression (CBR), gradient boosted regression tree (GBRT), support vector regression (SVR), and random forest regression (RFR) models. The XGBR-GA model yielded a promising result (R-2 = 0.683, RMSE = 25.08 Mgha(-1)) and outperformed the four other ML models. The XGBR-GA model retrieved a mangrove AGB ranging from 17 Mgha(-1) to 142 Mgha(-1) (with an average of 72.47 Mgha(-1)). Therefore, multisource optical and synthetic aperture radar (SAR) combined with the XGBR-GA model can be used to estimate the mangrove AGB in North Vietnam. The effectiveness of the proposed method needs to be further tested and compared to other mangrove ecosystems in the tropics.

关键词： industry 4.0   cyber-physical system   fault detection   random forest   long short-term memory   predictive maintenance   productivity improvement   artificial intelligence   big data and analytics   cyber-physical system design and operation   engineering informatics  

摘要： A cyber-physical system (CPS) is one of the key technologies of industry 4.0. It is an integrated system that merges computing, sensors, and actuators, controlled by computer-based algorithms that integrate people and cyberspace. However, CPS performance is limited by its computational complexity. Finding a way to implement CPS with reduced complexity while incorporating more efficient diagnostics, forecasting, and equipment health management in a real-time performance remains a challenge. Therefore, the study proposes an integrative machine-learning method to reduce the computational complexity and to improve the applicability as a virtual subsystem in the CPS environment. This study utilizes random forest (RF) and a time-series deep-learning model based on the long short-term memory (LSTM) networking to achieve real-time monitoring and to enable the faster corrective adjustment of machines. We propose a method in which a fault detection alarm is triggered well before a machine fails, enabling shop-floor engineers to adjust its parameters or perform maintenance to mitigate the impact of its shutdown. As demonstrated in two empirical studies, the proposed method outperforms other times-series techniques. Accuracy reaches 80% or higher 3 h prior to real-time shutdown in the first case, and a significant improvement in the life of the product (281%) during a particular process appears in the second case. The proposed method can be applied to other complex systems to boost the efficiency of machine utilization and productivity.

关键词： sensor-cloud   agricultural IoT   virtual sensor provisioning   machine learning   representative sensors  

摘要： The advent of sensor-cloud technology alleviates the limitations of traditional wireless sensor networks (WSNs) in terms of energy, storage, and computing, which has tremendous potential in various agricultural internet of things (IoT) applications. In the sensor-cloud environment, virtual sensor provisioning is an essential task. It chooses physical sensors to create virtual sensors in response to the users' requests. Considering the capricious meteorological environment of the outdoors, this paper presents an measurements similarity-based virtual-sensor provisioning scheme by taking advantage of machine learning in data analysis. First, to distinguish the changing trends, we classified all the physical sensors into several categories using historical data. Then, the k-means clustering algorithm was exploited for each class to cluster the physical sensors with high similarity. Finally, one representative physical sensor from each cluster was selected to create the corresponding virtual sensors. The experimental results show the reformation of our scheme with respect to energy efficiency, network lifetime, and data accuracy compared with the benchmark schemes.

关键词： Minced beef   Adulteration   Hyperspectral imaging   Texturized vegetable protein   Food fraud   Gluten   Pork  

摘要： The effectiveness of hyperspectral imaging (400-1000 nm) was proved as a nondestructive method to detect, classify, and quantify plant- and animal-based adulterants in minced beef and pork. Machine learning techniques were implemented to build classification and prediction models. Samples were first classified into adulterated (1 class) or pure (5 classes). The type of adulterant (6 classes) was then evaluated. Finally, the level of each adulterant was estimated using partial least squares regression. The optimal classification models based on selected wavelengths of test set yielded classification rates of 75-100% and 100% for pure and adulterated samples, respectively. Whereas, the rates were 83-100% depending on adulterant type. Prediction models for adulterant level yielded correlation coefficient, r, and ratio of performance to prediction, RPD, of 0.69(1.41) for beef adulterated with pork and textured vegetable protein (TVP), and 0.93(2.82) for beef adulterated with TVP. Improvement in results may be achieved with larger sample size.

关键词： Citrus HLB detection   UAV   Multispectral images   Machine learning  

摘要： Citrus Huanglongbing (HLB), also known as citrus greening, is the most destructive disease in the citrus industry. Detecting this disease as early as possible and eradicating the roots of HLB-infected trees can control its spread. Ground diagnosis is time-consuming and laborious. Large area monitoring method of citrus orchard with high accuracy is rare. This study evaluates the feasibility of large area detection of citrus HLB by low altitude remote sensing and commits to improve the accuracy of large-area detection. A commercial multispectral camera (ADC-lite) mounted on DJI M100 UAV(unmanned Aerial Vehicle) was used to collect green, red and near-infrared multispectral image of large area citrus orchard, a linear-stretch was performed to remove noise pixel, vegetation indices (VIs) were calculated followed by correlation analysis and feature compression using PCA (principal components analysis) and AutoEncoder to discover potential features. Several machine learning algorithms, such as support vector machine (SVM), k-nearest neighbour (kNN), logistic regression (LR), naive Bayes and ensemble learning, were compared to model the healthy and HLB-infected samples after parameter optimization. The results showed that the feature of PCA features of VIs combining with original DN (digital numbers) value generally have highest accuracy and agreement in all models, and the ensemble learning and neural network approaches had strong robustness and the best classification results (100% in AdaBoost and 97.28% in neural network) using threshold strategy.

关键词： RSW joint of multiple sheets   Ultrasonic test   Image processing   Static strength   Fatigue behavior   Artificial neural network   Genetic algorithm  

摘要： Ultrasonic Testing (UT) is one of the well-known Non-Destructive Techniques (NDT) of spot-weld inspection in the advanced industries, especially in automotive industry. However, the relationship between the UT results and strength of the spot-welded joints subjected to various loading conditions is unknown. The main purpose of this research is to present an integrated search system as a new approach for assessment of tensile strength and fatigue behavior of the spot-welded joints. To this end, Resistance Spot Weld (RSW) specimens of three-sheets were made of different types of low carbon steel. Afterward, the ultrasonic tests were carried out and the pulse-echo data of each sample were extracted utilizing Image Processing Technique (IPT). Several experiments (tensile and axial fatigue tests) were performed to study the mechanical properties of RSW joints of multiple sheets. The novel approach of the present research is to provide a new methodology for static strength and fatigue life assessment of three-sheets RSW joints based on the UT results by utilizing Artificial Neural Network (ANN) simulation. Next, Genetic Algorithm (GA) was used to optimize the structure of ANN. This approach helps to decrease the number of tests and the cost of performing destructive tests with appropriate

关键词： Data mining methods   Antihypertensive drugs   Drug-related attributes   Blood pressure control   Machine learning  

摘要： Globally, methods of controlling blood pressure in hypertension patients remain inefficient. The difficulty of prescribing appropriate drugs specific to a patient's clinical features serves as one of the most important factors. Characterizing the critical drug-related features, just like that of the antibacterial spectrum (where each item is sensitive to the targeted drug's effectiveness or a specified indication), may help a doctor easily prescribe appropriate drugs by matching a patient's attributes with drug-related features, and effectiveness of the selected drugs would also be ascertained. In this study, we aimed to apply data mining methods to obtain the clinical characteristics spectrum or important clinical features of five frequently used drugs (Irbesartan, Metoprolol, Felodipine, Amlodipine, and Levamlodipine) for hypertension control by comparing successful and unsuccessful cases. Spectrum analysis based on a statistical method and five algorithms based on machine learning were used to extract the critical clinical features. A visualized relative weight matrix was then achieved by combining the results from the characteristic spectrum and machine learning-based methods. Our results indicated that the five targeted antihypertension agents had different importance orders of the 15 relative clinical features. Clinical analysis showed that the extracted important clinical attributes of the five drugs were both reasonable and meaningful in the selection of hypertension treatment. Therefore, our study provided a data-driven reference for the personalization of clinical antihypertensive drugs.