关键词： Machine learning   Reconstructing missing data   Euphrates basin   Runoff prediction  

摘要： Machine learning (ML) techniques have been popular data-driven approaches for hydrological studies during the last few decades owing to their capability to identify complex nonlinear relationships between input and output data without the requirement for physical understanding of the system. This paper aims to predict river flows using various ML methods [feed forward neural network (FFNN), adaptive neuro fuzzy inference system (ANFIS), and genetic programming (GP)] and also a non-ML method (multiple linear regression) in the Euphrates Basin in Turkey. Infilling the missing data in the runoff record of the selected stations in Euphrates Basin is also an objective of this study. The ML methods were applied to the three main sub-basins of the Euphrates Basin, namely the Upper, Middle, and Lower Euphrates Basins. ANFIS and FFNN methods were the most successful ML methods for runoff estimation in the Upper and Lower Euphrates Basins, whereas GP and ANFIS models were the best ones in the Middle Euphrates Basin. Missing flow data were constructed successfully in the selected stations.

关键词： Landsat   ALOS/PALSAR   L-band   maximum likelihood classification   support vector machines   neural networks   random forest   topographic effects   Hyrcanian mountainous forest   Iran  

摘要： Forest environment classification in mountain regions based on single-sensor remote sensing approaches is hindered by forest complexity and topographic effects. Temperate broadleaf forests in western Asia such as the Hyrcanian forest in northern Iran have already suffered from intense anthropogenic activities. In those regions, forests mainly extend in rough terrain and comprise different stand structures, which are difficult to discriminate. This paper explores the joint analysis of Landsat7/ETM+, L-band SAR and their derived parameters and the effect of terrain corrections to overcome the challenges of discriminating forest stand age classes in mountain regions. We also verified the performances of three machine learning methods which have recently shown promising results using multisource data;support vector machines (SVM), neural networks (NN), random forest (RF) and one traditional classifier (i.e., maximum likelihood classification (MLC)) as a benchmark. The non-topographically corrected ETM+ data failed to differentiate among different forest stand age classes (average classification accuracy (OA) = 65%). This confirms the need to reduce relief effects prior data classification in mountain regions. SAR backscattering alone cannot properly differentiate among different forest stand age classes (OA = 62%). However, textures and PolSAR features are very efficient for the separation of forest classes (OA = 82%). The highest classification accuracy was achieved by the joint usage of SAR and ETM+ (OA = 86%). However, this shows a slight improvement compared to the ETM+ classification (OA = 84%). The machine learning classifiers proved t o be more robust and accurate compared to MLC. SVM and RF statistically produced better classification results than NN in the exploitation of the considered multi-source data.

关键词： Artificial Intelligence   Calibration   Computer Simulation   Groundwater   Water Movements   Water Movements  

摘要： Quantitative analyses of groundwater flow and transport typically rely on a physically-based model, which is inherently subject to error. Errors in model structure, parameter and data lead to both random and systematic error even in the output of a calibrated model. We develop complementary data-driven models (DDMs) to reduce the predictive error of physically-based groundwater models. Two machine learning techniques, the instance-based weighting and support vector regression, are used to build the DDMs. This approach is illustrated using two real-world case studies of the Republican River Compact Administration model and the Spokane Valley-Rathdrum Prairie model. The two groundwater models have different hydrogeologic settings, parameterization, and calibration methods. In the first case study, cluster analysis is introduced for data preprocessing to make the DDMs more robust and computationally efficient. The DDMs reduce the root-mean-square error (RMSE) of the temporal, spatial, and spatiotemporal prediction of piezometric head of the groundwater model by 82%, 60%, and 48%, respectively. In the second case study, the DDMs reduce the RMSE of the temporal prediction of piezometric head of the groundwater model by 77%. It is further demonstrated that the effectiveness of the DDMs depends on the existence and extent of the structure in the error of the physically-based model.

关键词： Genotype-phenotype relationship   Insertion/deletion   Machine learning   Indeuctive logic programming  

摘要： Background: Small insertion and deletion polymorphisms (Indels) are the second most common mutations in the human genome, after Single Nucleotide Polymorphisms (SNPs). Recent studies have shown that they have significant influence on genetic variation by altering human traits and can cause multiple human diseases. In particular, many Indels that occur in protein coding regions are known to impact the structure or function of the protein. A major challenge is to predict the effects of these Indels and to distinguish between deleterious and neutral variants. When an Indel occurs within a coding region, it can be either frameshifting (FS) or non-frameshifting (NFS). FS-Indels either modify the complete C-terminal region of the protein or result in premature termination of translation. NFS-Indels insert/delete multiples of three nucleotides leading to the insertion/deletion of one or more amino acids. Results: In order to study the relationships between NFS-Indels and Mendelian diseases, we characterized NFS-Indels according to numerous structural, functional and evolutionary parameters. We then used these parameters to identify specific characteristics of disease-causing and neutral NFS-Indels. Finally, we developed a new machine learning approach, KD4i, that can be used to predict the phenotypic effects of NFS-Indels. Conclusions: We demonstrate in a large-scale evaluation that the accuracy of KD4i is comparable to existing state-of-the- art methods. However, a major advantage of our approach is that we also provide the reasons for the predictions, in the form of a set of rules. The rules are interpretable by non-expert humans and they thus represent new knowledge about the relationships between the genotype and phenotypes of NFS-Indels and the causative molecular perturbations that result in the disease.

关键词： 电子商务交易量   站内搜索数据   商品分类  

摘要： 对搜索型商品(3C)、体验型商品(零食)、中间型商品(女装)分别建立基于淘宝网站内搜索数据对商品在线交易量的预测模型。研究结果表明,搜索数据对于搜索型商品的在线交易量具有最佳的预测效果,而对体验型商品的预测效果较差。

关键词： 数据挖掘   分类预测   敏感性   特异性  

摘要： 数据挖掘技术己经引起了信息产业界的广泛关注。分类预测是其中一个主要的研究方向,有着广泛的应用价值。对数据挖掘中的三种分类预测挖掘算法(C5.0、BP-人工神经网络和TAN贝叶斯网络)进行了研究和探讨,并通过一个实际例子对三种算法做了分析和比较,三种算法的精确度分别为87.26%、85.71%和88.7%。TAN贝叶斯网络的精确度和敏感性均最高,C5.0算法的特异性最好。

关键词： Distortion   Electrostatics   Force field   Kriging   Machine learning   Multipole moment   Protein Data Bank (PDB)   Quantum chemical topology   Serine  

摘要： A multipolar, polarizable electrostatic method for future use in a novel force field is described. Quantum Chemical Topology (QCT) is used to partition the electron density of a chemical system into atoms, then the machine learning method Kriging is used to build models that relate the multipole moments of the atoms to the positions of their surrounding nuclei. The pilot system serine is used to study both the influence of the level of theory and the set of data generator methods used. The latter consists of: (i) sampling of protein structures deposited in the Protein Data Bank (PDB), or (ii) normal mode distortion along either (a) Cartesian coordinates, or (b) redundant internal coordinates. Wavefunctions for the sampled geometries were obtained at the HF/6-31G(d,p), B3LYP/apc-1, and MP2/cc-pVDZ levels of theory, prior to calculation of the atomic multipole moments by volume integration. The average absolute error (over an independent test set of conformations) in the total atom-atom electrostatic interaction energy of serine, using Kriging models built with the three data generator methods is 11.3 kJ mol(-1) (PDB), 8.2 kJ mol(-1) (Cartesian distortion), and 10.1 kJ mol(-1) (redundant internal distortion) at the HF/6-31G(d, p) level. At the B3LYP/apc-1 level, the respective errors are 7.7 kJ mol(-1), 6.7 kJ mol(-1), and 4.9 kJmol(-1), while at the MP2/cc-pVDZ level they are 6.5 kJ mol(-1), 5.3 kJ mol(-1), and 4.0 kJmol(-1). The ranges of geometries generated by the redundant internal coordinate distortion and by extraction from the PDB are much wider than the range generated by Cartesian distortion. The atomic multipole moment and electrostatic interaction energy predictions for the B3LYP/apc-1 and MP2/cc-pVDZ levels are similar, and both are better than the corresponding predictions at the HF/6-31G(d, p) level.

关键词： ACTIVATED sludge process (Sewage purification)   ELECTROLYTIC oxidation   CHEMISTRY experiments   ORGANIC compounds removal (Sewage purification)   CHEMICAL oxygen demand   MACHINE learning  

摘要： In the present research, an electro-oxidation method was applied to decrease the organic compounds and remove the available micro-organisms in activated sludge of the sewage. Within this method, low cost electrodes were used, including stainless steel, graphite, and Pb/PbO2, and the operating parameters (pH, current density, and operating time) were experimentally optimized. In order to determine sludge stabilization (removal of organic matters and microorganisms), the decrease of parameters like chemical oxygen demand, the increase of electroconductivity and the total dissolved solids, total coli form, and fecal coli form were investigated. Two machine learning techniques (artificial neural networks and support vector machines) were applied comparatively for prediction of the process efficiency. Accurate results were obtained by simulation, in agreement with experimental data.

关键词： Structural interaction fingerprint   Machine learning   Virtual screening   Docking results analysis  

摘要： In this Letter, we present a novel methodology of searching for biologically active compounds, which is based on the combination of docking experiments and analysis of the results by machine learning methods. The study was performed for 5 different protein kinases, and several sets of compounds (active, inactive and assumed inactives) were docked into their targets. The resulting ligand-protein complexes were represented by the means of structural interaction fingerprints profiles (SIFts profiles) that constituted an input for ML methods. The developed protocol was found to be superior to the combination of classification algorithms with the standard fingerprint MACCSFP. (C) 2013 Elsevier Ltd. All rights reserved.

关键词： 软件产品   质量缺陷   预测性能   非均衡分类   随机森林算法   代价敏感  

摘要： 软件缺陷的存在不仅危害软件产品质量，还增加软件组织的管理成本，延缓了开发和发布进度。软件缺陷预测可以通过历史数据、已经发现的缺陷等表征软件相关产品、过程、资源特征的软件度量数据，来预测软件的缺陷数目和分布，是一种重要的缺陷预防和控制方式。目前软件缺陷预测的研究主要是从历史数据获取来源和预测方法这两方面入手。然而，获取到的软件历史缺陷数据往往是非均衡的，易错样本比例往往不足20％。传统的预测方法通常会以预测整体准确率为目标，而忽略少数类样本，进而给易错数据样本带来极大的误分率。因此，缺陷预测是一个非均衡分类问题。\n 非均衡分类问题是目前机器学习领域的一个热门课题，涵盖了从数据预处理角度和算法改进角度入手的两大研究方法。基于抽样的非均衡分类方法也有很多相关研究，然而对于最优的抽样方法和最优的分类算法，以及如何选择合适的抽样比例，都没有得到一个统一的结论，很多研究往往是单维度地考虑抽样方法或分类算法的优劣。\n 针对这一问题，本文提出使用经验研究的方法，在软件工程领域采用所提出的非均衡分类模型进行软件缺陷分类预测，多维度对比分析各类抽样方法、抽样比例和机器学习算法对缺陷预测中数据分布非均衡问题的解决能力。本文将该模型应用于真实的NASA数据集进行案例研究。通过经验性地系统化对比分析12种已有抽样与分类算法组合的预测性能优劣，得到SpreadSubsampling和随机森林结合的方法(SP-RF)综合表现最好，但具有较高伪正率(FPR)。\n 为了进一步提高预测性能，针对原始SP-RF方法会对原始数据带来较大的噪音及信息缺失等不足，从不同角度提出了两种基于SP-RF的改进算法，分别为内置均衡化抽样的自适应随机森林改进算法(IBSBA-RF)和基于样本分布加权投票的自适应随机森林算法(VBRF)。实验表明，IBSBA-RF算法可以显著降低预测结果的FPR，并且进一步提高了预测结果的AUC和Balance值。此外，本文还根据IBSBA-RF方法和VBRF方法设计并实现了支持基于非均衡分类模型的缺陷预测工具。