Lazy Learning Method Research Articles

Data learning‐based model‐free adaptive control and application to an NAO robot

AbstractIn this article, an improved model‐free adaptive control method based on the controller dynamic linearization technique combined with the locally weighted regression‐based lazy learning method (cMFAC‐LL) is presented, and it is applied to solve the path‐tracking control problem for an NAO robot. In the proposed cMFAC‐LL method, two dynamic linearization techniques are first applied on the controlled plant, and then the cMFAC‐LL controller is further designed with the time‐varying parameters estimated using a novel locally weighted regression‐based lazy learning (LWR‐LL) technique. The greatest advantage of the cMFAC‐LL method is that it is a pure data‐driven control method, and the designed controller makes full use of both the online and offline measurement data of the plant. Moreover, the introduction of the local learning (LL) method gives the cMFAC‐LL method strong data learning ability and satisfactory control performance. The stability of cMFAC‐LL is proven via rigorous mathematical analysis. Furthermore, cMFAC‐LL is applied to the path‐tracking control of the NAO robot under different walking environments, by which the applicability of cMFAC‐LL is further verified.

Optimizing resources allocation in a heterogeneous cloud radio access network using machine learning

AbstractIn this article, we propose an iterative algorithm based on lazy learning. The proposed scheme considers two consecutive algorithms for optimal resource allocation in heterogeneous cloud radio access networks (H‐CRAN). In the conventional approach, resources allocation is usually presented as a mathematical optimization problem and solved online using a real‐time scenario. Due to the non‐convex nature of resource allocation problems, finding optimal solutions in real‐time is a difficult challenge and has high computational complexity. To overcome these challenges in this paper, the problem is first formulated as a mixed‐integer nonlinear programming (MINLP) to optimize the system capacity and assign the underlay users to the subchannels, which is NP‐hard. To obtain the solution, in the first algorithm, the proposed problem is first reformulated as a convex problem using a matching theory and coalition formation method. Then, in the second algorithm, the lazy learning method is employed to find the optimal solution to the resources allocation problem. The lazy algorithm uses the output results of the first algorithm as training data, which are known as historical data. Similarities between current scenario data and historical scenario data, as well as the solution of resources allocation in historical scenarios, can be exploited to improve the resource allocation of the current scenario. Simulation results confirm that compared with the conventional approach, the suggested technique can significantly improve the system performance in terms of the total users assigned, overall throughput, the pace of resource allocation, and user fairness.

Usage The Lazy Learning Meta-Heuristic Technique for Predicting Entrepreneurial Marketing In The Insurance Industry

Due to the increasing importance of marketing, entrepreneurship and the role of organizational structure in their application, the purpose of this research is to predict entrepreneurial marketing using an organizational structure in the insurance industry. For this purpose, for marketing, seven indicators and for organizational structure, three indicators are defined, then prediction of entrepreneurial marketing indicators has been done by organizational structure indicators using lazy learning algorithm. In the proposed method, after predicting each data by K vector from its closest neighbor, the algorithm database is enriched for better prediction of future data. The proposed algorithm is simulated and compared in five different modes by MATLAB software, also, three insurance (Iran, Karafarin and Parsiyan) companies are selected in Mazandaran province. In total, the statistical population in this study is 588 cases. The results of simulation indicate the proper accuracy of entrepreneurial marketing forecasting based on validation parameters MSE and NRMSD. In this research, Lazy Learning method can predict future without modeling the problem with previous information processing.

Data Mining Reconstruction of Magnetotail Reconnection and Implications for Its First-Principle Modeling

Magnetic reconnection is a fundamental process providing topological changes of the magnetic field, reconfiguration of space plasmas and release of energy in key space weather phenomena, solar flares, coronal mass ejections and magnetospheric substorms. Its multiscale nature is difficult to study in observations because of their sparsity. Here we show how the lazy learning method, known as K nearest neighbors, helps mine data in historical space magnetometer records to provide empirical reconstructions of reconnection in the Earth’s magnetotail where the energy of solar wind-magnetosphere interaction is stored and released during substorms. Data mining reveals two reconnection regions (X-lines) with different properties. In the mid tail (∼30RE from Earth, where RE is the Earth’s radius) reconnection is steady, whereas closer to Earth (∼20RE) it is transient. It is found that a similar combination of the steady and transient reconnection processes can be reproduced in kinetic particle-in-cell simulations of the magnetotail current sheet.

A New Hybrid Instance-Based Learning Model for Decision-Making in the P2P Lending Market

Peer-to-Peer (P2P) lending has grown rapidly in the past years. Therefore, borrowers and lenders are provided with the opportunity of lending and borrowing independently of the banks. Lenders in the P2P lending market can share their total investment amount among different loans, so making a decision may be difficult for inexpert lenders. The aim of this study is to propose a novel decision-making framework in which instance-based learning as a lazy learning method and artificial neural networks as an eager learning approach are integrated. The proposed hybrid instance-based learning (HIBL) model has the ability to predict the return and risk of new loans and help investors to find the optimal portfolio. In order to check the effectiveness of our model, we use a real-world dataset from one of the most popular P2P lending marketplaces, namely Lending Club. Moreover, a comparison among our proposed model and a rating-based method reveals that the proposed HIBL model can improve investments in P2P lending.

Memory Network for Linguistic Structure Parsing

Memory-based learning can be characterized as a lazy learning method in machine learning terminology because it delays the processing of input by storing the input until needed. Linguistic structure parsing, which has been in a performance improvement bottleneck since the latest series of works was presented, determines the syntactic or semantic structure of a sentence. In this article, we construct a memory component and use it to augment a linguistic structure parser which allows the parser to directly extract patterns from the known training treebank to form memory. The experimental results show that existing state-of-the-art parsers reach new heights of performance on the main benchmarks for dependency parsing and semantic role labeling with this memory network.

Analysis of algorithm support vector machine learning and k-nearest neighbor in data accuracy

K-Nearest Neighbor is a method of lazy learning method which is a group of instances-based learning. K-NN searches by searching for groups of objects in the training data that are closest to the object on new data or testing data. Support Vector Machine is a learning machine method that works with the aim of finding the best hyperplane that separates two classes in input space. School Achievement is an achievement obtained by serious learning and discipline. The category of outstanding students is to get a good average score and not have an attendance list, especially Absent (A) and a list of late attendance at school can be classified to obtain information on the accuracy of the data being tested. In the testing process both methods obtained good accuracy results between the two methods, namely K-NN obtained an accuracy of 88.52% while SVM is 91.07%.

AEkNN: An AutoEncoder kNN—Based Classifier With Built-in Dimensionality Reduction

High dimensionality tends to be a challenge for most machine learning tasks, including classification. There are different classification methodologies, of which instance-based learning is one. One of the best known members of this family is the k-nearest neighbors (kNNs) algorithm. Its strategy relies on searching a set of nearest instances. In high-dimensional spaces, the distances between examples lose significance. Therefore, kNN, in the same way as many other classifiers, tends to worsen its performance as the number of input variables grows. In this study, AEkNN, a new kNN-based algorithm with built-in dimensionality reduction, is presented. Aiming to obtain a new representation of the data, having a lower dimensionality but with more informational features, AEkNN internally uses autoencoders. From this new vector of features the computed distances should be more significant, thus providing a way to choose better neighbors. An experimental evaluation of the new proposal is conducted, analyzing several configurations and comparing them against the original kNN algorithm and classical dimensionality reduction methods. The obtained conclusions demonstrate that AEkNN offers better results in predictive and runtime performance.

In Silico Prediction of Chemical Toxicity Profile Using Local Lazy Learning

Chemical toxicity is an important reason for late-stage failure in drug R&D. However, it is time-consuming and expensive to identify the multiple toxicities of compounds using the traditional experiments. Thus, it is attractive to build an accurate prediction model for the toxicity profile of compounds. In this study, we carried out a research on six types of toxicities: (I) Acute Toxicity; (II) Mutagenicity; (III) Tumorigenicity; (IV) Skin and Eye Irritation; (V) Reproductive Effects; (VI) Multiple Dose Effects, using local lazy learning (LLL) method for multi-label learning. 17,120 compounds were split into the training set and the test set as a ratio of 4:1 by using the Kennard-Stone algorithm. Four types of properties, including molecular fingerprints (ECFP_4 and FCFP_4), descriptors, and chemical-chemical-interactions, were adopted for model building. The model 'ECFP_4+LLL' yielded the best performance for the test set, while balanced accuracy (BACC) reached 0.692, 0.691, 0.666, 0.680, 0.631, 0.599 for six types of toxicities, respectively. Furthermore, some essential toxicophores for six types of toxicities were identified by using the Laplacian-modified Bayesian model. The accurate prediction model and the chemical toxicophores can provide some guidance for designing drugs with low toxicity.

The influence of alternative data smoothing prediction techniques on the performance of a two-stage short-term urban travel time prediction framework

ABSTRACTThis article investigates the impact of alternative data smoothing and traffic prediction methods on the accuracy of the performance of a two-stage short-term urban travel time prediction framework. Using this framework, we test the influence of the combination of two different data smoothing and four different prediction methods using travel time data from two substantially different urban traffic environments and under both normal and abnormal conditions. This constitutes the most comprehensive empirical evaluation of the joint influence of smoothing and predictor choice to date. The results indicate that the use of data smoothing improves prediction accuracy regardless of the prediction method used and that this is true in different traffic environments and during both normal and abnormal (incident) conditions. Moreover, the use of data smoothing in general has a much greater influence on prediction performance than the choice of specific prediction method, and this is independent of the specific smoothing method used. In normal traffic conditions, the different prediction methods produce broadly similar results but under abnormal conditions, lazy learning methods emerge as superior.

Multi-label classification algorithm research based on swarm intelligence

Since data and resources have massive feature and feature of data are increasingly complex, traditional data structures are not suitable for current data anymore. Therefore, traditional single-label learning method cannot meet the requirements of technology development and the importance of multi-label leaning method becomes more and more highlighted. K-Nearest Neighbor (KNN) classification method is a lazy learning method in data classification methods. It does not need data training process and theoretical system is mature. In addition, principle and implementation is simple. This paper proposed improvements strategies only considers numerical feature of sample KNN when classifying, but not consider the disadvantage of sample structure feature. This paper introduced particle swarm optimization algorithm into KNN classification and make adjustments to Euclidean distance formula in traditional KNN classification algorithm and add weight value to each feature. Using adjusted distance formula to train training data through particle swarm optimization algorithm and optimized a set of weight value for all features and put these optimized weight values to adjusted distance formula and calculated the distance between each example in test data set and in training data set and predict the test data set. Experiment results show that weighted KNN classification algorithm based on particle swarm optimization algorithm can achieve better classification accuracy than traditional KNN classification algorithm.

Short-term load forecasting using mixed lazy learning method

A novel short-term load forecasting method based on the lazy learning (LL) algorithm is proposed. The LL algorithm's input data are electrical load information, daily electricity consumption patterns, and temperatures in a specified region. In order to verify the ability of the proposed method, a load forecasting problem, using the Pennsylvania-New Jersey-Maryland Interconnection electrical load data, is carried out. Three LL models are proposed: constant, linear, and mixed models. First, the performances of the 3 developed models are compared using the root mean square error technique. The best technique is then selected to compete with the state-of-the-art neural network (NN) load forecasting models. A comparison is made between the performances of the proposed mixed-model LL as the superior LL model and the radial basis function and multilayer perceptron NN models. The results reveal significant improvements in the precision and efficiency of the proposed forecasting model when compared with the NN techniques.

Estimation of acute oral toxicity in rat using local lazy learning.

BackgroundAcute toxicity means the ability of a substance to cause adverse effects within a short period following dosing or exposure, which is usually the first step in the toxicological investigations of unknown substances. The median lethal dose, LD50, is frequently used as a general indicator of a substance’s acute toxicity, and there is a high demand on developing non-animal-based prediction of LD50. Unfortunately, it is difficult to accurately predict compound LD50 using a single QSAR model, because the acute toxicity may involve complex mechanisms and multiple biochemical processes.ResultsIn this study, we reported the use of local lazy learning (LLL) methods, which could capture subtle local structure-toxicity relationships around each query compound, to develop LD50 prediction models: (a) local lazy regression (LLR): a linear regression model built using k neighbors; (b) SA: the arithmetical mean of the activities of k nearest neighbors; (c) SR: the weighted mean of the activities of k nearest neighbors; (d) GP: the projection point of the compound on the line defined by its two nearest neighbors. We defined the applicability domain (AD) to decide to what an extent and under what circumstances the prediction is reliable. In the end, we developed a consensus model based on the predicted values of individual LLL models, yielding correlation coefficients R2 of 0.712 on a test set containing 2,896 compounds.ConclusionEncouraged by the promising results, we expect that our consensus LLL model of LD50 would become a useful tool for predicting acute toxicity. All models developed in this study are available via http://www.dddc.ac.cn/admetus.

ESTIMATING PHOTOMETRIC REDSHIFTS OF QUASARS VIA THEk-NEAREST NEIGHBOR APPROACH BASED ON LARGE SURVEY DATABASES

We apply one of lazy learning methods named k-nearest neighbor algorithm (kNN) to estimate the photometric redshifts of quasars, based on various datasets from the Sloan Digital Sky Survey (SDSS), UKIRT Infrared Deep Sky Survey (UKIDSS) and Wide-field Infrared Survey Explorer (WISE) (the SDSS sample, the SDSS-UKIDSS sample, the SDSS-WISE sample and the SDSS-UKIDSS-WISE sample). The influence of the k value and different input patterns on the performance of kNN is discussed. kNN arrives at the best performance when k is different with a special input pattern for a special dataset. The best result belongs to the SDSS-UKIDSS-WISE sample. The experimental results show that generally the more information from more bands, the better performance of photometric redshift estimation with kNN. The results also demonstrate that kNN using multiband data can effectively solve the catastrophic failure of photometric redshift estimation, which is met by many machine learning methods. By comparing the performance of various methods for photometric redshift estimation of quasars, kNN based on KD-Tree shows its superiority with the best accuracy for our case.

Opportunistic channel selection based on time series prediction in cognitive radio networks

ABSTRACTTo improve the utilisation efficiency of licensed spectrum in cognitive radio network, two new channel selection strategies are proposed for the secondary user (SU) in this paper. The proposed solution tries its best to reduce collision and switching probabilities of the SU during data transmission. By using historical information of the licensed spectrum, the SU chooses the channel with the lowest busy probability within its service time for data transmission. Time series prediction is employed to forecast the near future busy probabilities of the licensed spectrum units, and a novel time series prediction method named distance factor recursive least square is also presented. Simulations prove that the performances of the SU, which is measured by collision probability with primary user, switching probability during data transmission and throughput within limited time slots, are all significantly improved when compared with random channel selection method. Higher prediction accuracy than the conventional recursive least square and lazy learning methods is achieved by proposed time series prediction algorithm when tested by voice traffic data and Lorenz time series. Copyright © 2013 John Wiley & Sons, Ltd.

A generalized cluster centroid based classifier for text categorization

In this paper, a Generalized Cluster Centroid based Classifier (GCCC) and its variants for text categorization are proposed by utilizing a clustering algorithm to integrate two well-known classifiers, i.e., the K-nearest-neighbor (KNN) classifier and the Rocchio classifier. KNN, a lazy learning method, suffers from inefficiency in online categorization while achieving remarkable effectiveness. Rocchio, which has efficient categorization performance, fails to obtain an expressive categorization model due to its inherent linear separability assumption. Our proposed method mainly focuses on two points: one point is that we use a clustering algorithm to strengthen the expressiveness of the Rocchio model; another one is that we employ the improved Rocchio model to speed up the categorization process of KNN. Extensive experiments conducted on both English and Chinese corpora show that GCCC and its variants have better categorization ability than some state-of-the-art classifiers, i.e., Rocchio, KNN and Support Vector Machine (SVM).

A hybrid text classification approach with low dependency on parameter by integrating K-nearest neighbor and support vector machine

This work implements a new text document classifier by integrating the K-nearest neighbor (KNN) classification approach with the support vector machine (SVM) training algorithm. The proposed Nearest Neighbor-Support Vector Machine hybrid classification approach is coined as SVM-NN. The KNN has been reported as one of the widely used text classification approaches due to its simplicity and efficiency in handling various types of text classification tasks. However, there exists a major problem of the KNN in determining the appropriate value for parameter K in order to guarantee high classification effectiveness. This is due to the fact that the selection of the value of parameter K has high impact on the accuracy of the KNN classifier. Other than determining the optimal value of parameter K, the KNN is also a lazy learning method which keeps the entire training samples until classification time. Hence, the computational process of the KNN has become intensive when the value of parameter K increases. In this paper, we propose the SVM-NN hybrid classification approach with the objective that to minimize the impact of parameter on classification accuracy. In the training stage, the SVM is utilized to reduce the training samples for each of the available categories to their support vectors (SVs). The SVs from different categories are used as the training data of nearest neighbor classification algorithm in which the Euclidean distance function is used to calculate the average distance between the testing data point to each set of SVs of different categories. The classification decision is made based on the category which has the shortest average distance between its SVs and the testing data point. The experiments on several benchmark text datasets show that the classification accuracy of the SVM-NN approach has low impact on the value of parameter, as compared to the conventional KNN classification model.

Similarity measures over refinement graphs

Similarity also plays a crucial role in support vector machines. Similarity assessment plays a key role in lazy learning methods such as k-nearest neighbor or case-based reasoning. In this paper we will show how refinement graphs, that were originally introduced for inductive learning, can be employed to assess and reason about similarity. We will define and analyze two similarity measures, S ? and S ? , based on refinement graphs. The anti-unification-based similarity, S ? , assesses similarity by finding the anti-unification of two instances, which is a description capturing all the information common to these two instances. The property-based similarity, S ? , is based on a process of disintegrating the instances into a set of properties, and then analyzing these property sets. Moreover these similarity measures are applicable to any representation language for which a refinement graph that satisfies the requirements we identify can be defined. Specifically, we present a refinement graph for feature terms, in which several languages of increasing expressiveness can be defined. The similarity measures are empirically evaluated on relational data sets belonging to languages of different expressiveness.

Neural-empirical tyre model based on recursive lazy learning under combined longitudinal and lateral slip conditions

The behaviour of the tyre plays an important role in the vehicle handling. An accurate tyre model that estimates these forces and moments it is highly essential for the studies of vehicle behaviour. For the last ten years neural networks have attracted a great deal of attention in vehicle dynamics and control. Neural networks have been effectively applied to model complex systems due to their good learning capability. In this paper a recursive lazy learning method based on neural networks is considered to model the tyre characteristics under combined braking and cornering. The proposed method is validated by comparison with experimental obtained responses. Results show the estimated model correlates very well with the data obtained experimentally. Moreover, the neural model proposed allows to include the asymetric tyre behaviour in the tyre model without difficulty.

An FPGA-implemented Associative-memory Based Online Learning Method

In this work, we propose an FPGA implemented associative-memory-based lazy learning method with a short/long-term memory concept which has an online learning capability. The conventional lazy or memory-based learning algorithm has low computational costs in the training stage since it simply stores the inputs and defers processing of training data until a query needs to be answered. However, it requires large storage capability, often pays high computational costs to answer a request, and is easily fooled by irrelevant attributes. To speed-up the query time, which is a deficiency of conventional lazy learning, we apply an associative memory which is implemented in an FPGA for searching the most similar data among the previously stored reference data in parallel. The proposed learning method can eliminate irrelevant attributes and reduce the storage requirements through a forgetting process in the short-term memory. When the online lazy learning model is applied to handwritten character recognition, the mismatch rate is reduced to 0.6% after a learning process with 450 alphabet letters learning and the number of reference is reduced by about 40% in comparison to the conventional method.