Abstract: mellitus or simply diabetes is a disease caused due to the increase level of blood glucose. Various available traditional methods for diagnosing diabetes are based on physical and chemical tests. These methods can have errors due to different uncertainties. A number of Data mining algorithms were designed to overcome these uncertainties. Among these algorithms, amalgam KNN and ANFIS provides higher classification accuracy than the existing approaches. The main data mining algorithms discussed in this paper are EM algorithm, KNN algorithm, K-means algorithm, amalgam KNN algorithm and ANFIS algorithm. EM algorithm is the expectation-maximization algorithm used for sampling, to determine and maximize the expectation in successive iteration cycles. KNN algorithm is used for classifying the objects and used to predict the labels based on some closest training examples in the feature space. K means algorithm follows partitioning methods based on some input parameters on the datasets of n objects. Amalgam combines both the features of KNN and K means with some additional processing. ANFIS is the Adaptive Neuro Fuzzy Inference System which combines the features of adaptive neural network and Fuzzy Inference System. The data set chosen for classification and experimental simulation is based on Pima Indian Diabetic Set from University of California, Irvine (UCI) Repository of Machine Learning databases. Keywordsmining, Diabetes, EM algorithm, KNN algorithm, K- means algorithm, amalgam KNN algorithm, ANFIS algorithm

Abstract: The Travelling Salesman Problem (TSP) has been extensively studied in the literature and various solvers are available. However, none of the state-of-the-art solvers for TSP outperforms the others in all problem instances within a given time limit. Therefore, the prediction of the best performing algorithm can save computational resources and optimise the results. In this paper, the TSP is studied in context of automated algorithm selection. Our aim is to identify the relevant features of problem instances and tackle this scenario as a machine learning task. We extend the set of existing features in the literature and propose several novel features to better characterise the problem. The contribution of the new features is statistically analysed and experiments show that adding our new features improves the prediction accuracy. We identified that our features based on kNN graph transformation are especially helpful. To create the training datasets, two state-of-the-art (meta-) heuristic algorithms are systematically evaluated on more than 2000 problems. Overall, we show that our prediction can be substantially more accurate than simple preference of an algorithm with the best performance for a majority of problem instances.

Abstract: Matching is an important combinatorial problem with a number of applications in areas such as community detection, sparse linear algebra, and network alignment. Since computing optimal matchings can be very time consuming, several fast approximation algorithms, both sequential and parallel, have been suggested. Common to the algorithms giving the best solutions is that they tend to be sequential by nature, while algorithms more suitable for parallel computation give solutions of lower quality. We present a new simple 1/2-approximation algorithm for the weighted matching problem. This algorithm is both faster than any other suggested sequential 1/2-approximation algorithm on almost all inputs and when parallelized also scales better than previous multithreaded algorithms. We further extend this to a general scalable multithreaded algorithm that computes matchings of weight comparable with the best sequential deterministic algorithms. The performance of the suggested algorithms is documented through extensive experiments on different multithreaded architectures.

Abstract: The k-means algorithm is well-known for its efficiency in clustering large data sets and it is restricted to the numerical data types. But the real world is a mixture of various data typed objects. In this paper we implemented algorithms which extend the k-means algorithm to categorical domains by using Modified k-modes algorithm and domains with mixed categorical and numerical values by using k-prototypes algorithm. The Modified k-modes algorithm will replace the means with the modes of the clusters by following three measures like “using a simple matching dissimilarity measure for categorical data”, “replacing means of clusters by modes” and “using a frequency-based method to find the modes of a problem used by the k-means algorithm”. The other algorithm used in this paper is the k-prototypes algorithm which is implemented by integrating the Incremental k-means and the Modified k-modes partition clustering algorithms. All these algorithms reduce the cost function value.

Abstract: A new text classification algorithm has been put forward based on basic support vector machine algorithm. The SVM-KNN algorithm for text classification has been proposed which combined SVM algorithm and KNN algorithm. The SVM-KNN algorithm can improve the performance of classifier by the feedback and improvement of classifying prediction probability. The actual effect of SVM-KNN algorithm is tested and the performance is proved in related Chinese web page classification test system.

Abstract: Sentiment classification is to find the polarity of product or user reviews. Supervised machine learning algorithms are used for opinion mining such as Navie Bayes, K-nearest neighbor and Support vector machine. KNN is simple algorithm but less efficient classification algorithm. In this paper we propose an improved KNN algorithm, genetic algorithm is developed which is a hybrid genetic algorithm that incorporates the information gain for feature selection and combined with KNN to improve its classification performance. Specifically, we compared other supervised machine learning approaches such as Navie Bayes and traditional kNN for Sentiment Classification of movie reviews and book reviews. The experimental results using genetic algorithm with improved indicate high performance levels with Fmeasure of over 87% on the movie reviews.

Abstract: The success of machine learning on a given task dependson, among other things, which learning algorithm is selected and its associated hyperparameters. Selecting an appropriate learning algorithm and setting its hyperparameters for a given data set can be a challenging task, especially for users who are not experts in machine learning. Previous work has examined using meta-features to predict which learning algorithm and hyperparameters should be used. However, choosing a set of meta-features that are predictive of algorithm performance is difficult. Here, we propose to apply collaborative filtering techniques to learning algorithm and hyperparameter selection, and find that doing so avoids determining which meta-features to use and outperforms traditional meta-learning approaches in many cases.

Abstract: In this paper, we study the problem of regenerator site (RS) selection for mixed line rate optical networks (MLR-RSS), with the objective of minimizing the number of RSs for a given set of requests. We first provide the problem definition of MLR-RSS and show that the MLR-RSS problem is NP-complete. An integer linear programming model is formulated. We then present two heuristic algorithms, named the independent algorithm and the sequential algorithm, and two approximation algorithms, named the MLR-combined algorithm and the weighted MLR-combined algorithm. The performance of the algorithms is compared via simulation, and results show that the weighted MLR-combined algorithm has the best performance. Results suggest that our proposed MLR algorithm outperforms existing single line rate algorithms by more than 20%. Also, the RS distribution suggests that certain nodes in the network have a much higher probability of being chosen as RSs than the others.

Abstract: Association Rule mining is one of the most important fields in data mining and knowledge discovery. This paper proposes an algorithm that combines the simple association rules derived from basic Apriori Algorithm with the multiple minimum support using maximum constraints. The algorithm is implemented, and is compared to its predecessor algorithms using a novel proposed comparison algorithm. Results of applying the proposed algorithm show faster performance than other algorithms without scarifying the accuracy.

Abstract: The amount of students data in the education system databases is growing day by day, so the knowledge taken out from these data need to be updated continuously. The training set of the supervised learning algorithms contains student's score in the test. Incremental learning ability is further significant for machine learning methodologies as student's data and the information is increasing. Against to the classical batch learning algorithm, incremental learning algorithm tries to forget unrelated information while training new instances. Now a days, combination of a classifiers is a novel concept for overall progress in the classification result. Therefore, an incremental ensemble of two classifiers namely Naive Bayes, K-Star using majority voting scheme is proposed. The large scale comparison of a proposed ensemble technique by using different voting scheme with the state-of the art algorithm on the student's data set has been done. The experimental results shown high accuracy for the proposed ensemble for the student's classification. High accuracy was also achieved for the majority voting scheme as compared to other voting scheme.

Abstract: The results from most machine learning experiments are used for a specific purpose and then discarded. This results in a significant loss of information and requires rerunning experiments to compare learning algorithms. This also requires implementation of another algorithm for comparison, that may not always be correctly implemented. By storing the results from previous experiments, machine learning algorithms can be compared easily and the knowledge gained from them can be used to improve their performance. The purpose of this work is to provide easy access to previous experimental results for learning and comparison. These stored results are comprehensive -- storing the prediction for each test instance as well as the learning algorithm, hyperparameters, and training set that were used. Previous results are particularly important for meta-learning, which, in a broad sense, is the process of learning from previous machine learning results such that the learning process is improved. While other experiment databases do exist, one of our focuses is on easy access to the data. We provide meta-learning data sets that are ready to be downloaded for meta-learning experiments. In addition, queries to the underlying database can be made if specific information is desired. We also differ from previous experiment databases in that our databases is designed at the instance level, where an instance is an example in a data set. We store the predictions of a learning algorithm trained on a specific training set for each instance in the test set. Data set level information can then be obtained by aggregating the results from the instances. The instance level information can be used for many tasks such as determining the diversity of a classifier or algorithmically determining the optimal subset of training instances for a learning algorithm.

Abstract: Weighted majority votes allow one to combine the output of several classifiers or voters. MinCq is a recent algorithm for optimizing the weight of each voter based on the minimization of a theoretical bound over the risk of the vote with elegant PAC-Bayesian generalization guarantees. However, while it has demonstrated good performance when combining weak classifiers, MinCq cannot make use of the useful a priori knowledge that one may have when using a mixture of weak and strong voters. In this paper, we propose P-MinCq, an extension of MinCq that can incorporate such knowledge in the form of a constraint over the distribution of the weights, along with general proofs of convergence that stand in the sample compression setting for data-dependent voters. The approach is applied to a vote of $$k$$k-NN classifiers with a specific modeling of the voters' performance. P-MinCq significantly outperforms the classic $$k$$k-NN classifier, a symmetric NN and MinCq using the same voters. We show that it is also competitive with LMNN, a popular metric learning algorithm, and that combining both approaches further reduces the error.

Abstract: In order to improve the accuracy of multi-factor classification algorithm, according to the relationship between the inclusion relation of sets and conceptual reasoning, this paper put forward a new algorithm for mining knowledge which is different from the decision tree algorithm, introduced the concepts of Or operation, Background Space, Decided Domain, Decided Degree, and Advantage factor, and presented the principle and concept of the knowledge mining based on mathematic description This paper also studied the problem of the training and test of the algorithm Training and testing of the algorithm was conducted on the breast cancer data set in the UCI shared database, and the total error rate is lower than the results of the see5 The result of study shows that the algorithm is simple in principle, better in learning capability and concise in knowledge expressing style

Abstract: One of the most widely used algorithms to solve clustering problems is the K-means. Despite of the algorithm's timely performance to find a fairly good solution, it shows some drawbacks like its dependence on initial conditions and trapping in local minima. This paper proposes a novel hybrid algorithm, comprised of K-means and a variation operator inspired by mutation in evolutionary algorithms, called Noisy K-means Algorithm (NKA). Previous research used K-means as one of the genetic operators in Genetic Algorithms. However, the proposed NKA is a kind of individual based algorithm that combines advantages of both K-means and mutation. As a result, proposed NKA algorithm has the advantage of faster convergence time, while escaping from local optima. In this algorithm, a probability function is utilized which adaptively tunes the rate of mutation. Furthermore, a special mutation operator is used to guide the search process according to the algorithm performance. Finally, the proposed algorithm is compared with the classical K-means, SOM Neural Network, Tabu Search and Genetic Algorithm in a given set of data. Simulation results statistically demonstrate that NKA out-performs all others and it is prominently prone to real time clustering.

Abstract: We propose a new simple and natural algorithm for learning the optimal Q-value function of a discounted-cost Markov Decision Process (MDP) when the transition kernels are unknown. Unlike the classical learning algorithms for MDPs, such as Q-learning and actor-critic algorithms, this algorithm doesn't depend on a stochastic approximation-based method. We show that our algorithm, which we call the empirical Q-value iteration (EQVI) algorithm, converges to the optimal Q-value function. We also give a rate of convergence or a non-asymptotic sample complexity bound, and also show that an asynchronous (or online) version of the algorithm will also work. Preliminary experimental results suggest a faster rate of convergence to a ball park estimate for our algorithm compared to stochastic approximation-based algorithms.

Abstract: Knowledge about algorithm similarity is an important aspect of meta-learning, where the information gathered from previous learning tasks can be used to guide the selection of algorithms for new datasets. Usually this task is done by comparing global performance measures across different datasets or alternatively, comparing the performance of algorithms at the instance-level. In both cases, the previous similarity measures do not consider misclassification costs, and hence they neglect an important information that can be exploited in different learning contexts. In this paper we present algorithm similarity measures that deals with cost proportions and different threshold choice methods for building crisp classifiers from learned models. Experiments were performed in a meta-learning study with 50 different learning tasks. The similarity measures were adopted to cluster algorithms according to their aggregated performance on the learning tasks. The clustering process revealed similarity between algorithms under different perspectives.

Abstract: Feature Selection FS has become the motivation of much research on decision support systems areas for which datasets with large number of features are analyzed. This paper presents a new method for the diagnosis of Coronary Artery Diseases CAD founded on Genetic Algorithm GA wrapper Bayes Naive BN. Initially, thirteen attributes were involved in predicting CAD. In GA-BN algorithm, GA produces in each iteration a subset of attributes that will be evaluated using the BN in the second step of the selection procedure. The final result set of attribute holds the most pertinent feature model that increases the accuracy. The accuracy results showed that the algorithm produces 85.50% classification accuracy in the diagnosis of CAD. Therefore, the strength of the Algorithm is then compared with other machine learning algorithms such as Support Vector Machine SVM, Multi-Layer Perceptron MLP and C4.5 decision tree Algorithm. The result of classification accuracy for those algorithms are respectively 83.5%, 83.16% and 80.85%. Then, the GA wrapper BN Algorithm is similarly compared with other FS algorithms. The Obtained results have shown very favorable outcomes for the diagnosis of CAD.

Abstract: Social network is a collection of heterogeneous multi-relational data represented by the graph, whose nodes represent object, whose edges represent relationships between nodes, and the weights represent the extent of the relationship between nodes. This paper gave a weighted K-means algorithm and introduced weighted K-means algorithm into social networks. Traditional k-means and most k-means variants are still computationally expensive for large datasets, however, the weighted K-means algorithm is to reduce the initial cluster centers blindness and randomness by eliminating noise point and narrowing the range of k values. Experiments datasets show that the weighted K-means algorithm significantly enhances the clustering quality. Therefore, the weighted K-means algorithm is effective and suitable for the social network. Algorithm’s error rate is smaller and accuracy is higher than that of traditional k-means algorithm.

Abstract: Trust models are mechanisms that allow agents to build trust without relying on a trusted central authority. Our goal was to develop a trust model that would operate with values that humans easily understand and manipulate: qualitative and ordinal values. The result is a trust model that computes trust from experiences created in interactions and from opinions obtained from third-party agents. The trust model, termed qualitative trust model (QTM), uses qualitative and ordinal values for assessing experiences, expressing opinions and estimating trust. We treat such values appropriately; we never convert them to numbers, but merely use their relative order. To aggregate a collection of such values, we propose an aggregation method that is based on comparing distributions and show some of its properties; the method can be used in other domains and can be seen as an alternative to median and similar methods. To cope with lying agents, QTM estimates trustworthiness in opinion providers with a modified version of the weighted majority algorithm, and additionally combines trustworthiness with social links between agents; such links are obtained implicitly by observing how agents provide opinions about each other. Finally, we compare QTM against a set of well-known trust models and demonstrate that it consistently performs well and on par with other quantitative models, and in many cases even outperforms them, particularly when the number of direct experiences is low.

Abstract: In this technical report we presented a novel approach to machine learning. Once the new framework is presented, we will provide a simple and yet very powerful learning algorithm which will be benchmark on various dataset. The framework we proposed is based on booleen circuits; more specifically the classifier produced by our algorithm have that form. Using bits and boolean gates instead of real numbers and multiplication enable the the learning algorithm and classifier to use very efficient boolean vector operations. This enable both the learning algorithm and classifier to be extremely efficient. The accuracy of the classifier we obtain with our framework compares very favorably those produced by conventional techniques, both in terms of efficiency and accuracy.

Abstract: The mainstream of current link prediction algorithm based on network topology structure generally have the problem of low efficiency of recalls. Study found that the correct results from some of the link prediction algorithms are complementary, accordingly, the Boosting method was considered to improve it. According to whether there is a link relationship between the nodes, the problem was divided into two categories, thus the link prediction algorithm as a two classification problem was defined. Furthermore, the algorithm complementary principle to select a number of representative link prediction algorithms as weak classifiers was followed, and a novel link prediction algorithm based on the AdaBoost algorithm was come up. The experimental results on the data from real dataset like the arXiv paper cooperation network and E-mail network show that, the novel algorithm has a better accuracy than the current mainstream algorithms.

Abstract: In order to deal with the problem of low categorization accuracy of minority class of the uneven forestry information text classification algorithm, this paper puts forward the uneven forestry information text classification algorithm based on weighted Gauss-Newton neural network, on the basis of weighted Gauss-Newton algorithm, the algorithm is proved via singular value decomposition principle. The experimental result shows that the algorithm has higher classification accuracy of majority class and minority class than algorithm of common classification. The algorithm expands a new method for the research on the uneven forestry information text classification algorithm. DOI : http://dx.doi.org/10.11591/telkomnika.v12i5.4388

Abstract: An algorithm of weighted k-means clustering is improved in this paper, which is based on improved genetic algorithm. The importance of different contributors in the process of manufacture is not the same when clustering, so the weight values of the parameters are considered. Retaining the best individuals and roulette are combined to decide which individuals are chose to crossover or mutation. Dynamic mutation operators are used here to decrease the speed of convergence. Two groups of data are used to make comparisons among the three algorithms, which suggest that the algorithm has overcome the problems of local optimum and low speed of convergence. The results show that it has a better clustering.

Abstract: The main drawback of reinforcement learning is that it learns nothing from an episode until it is over. So the learning procedure is very slow in case of large space applications. Differential Evolution (DE) algorithm is a population-based evolutionary optimization algorithm able to learn the search space in iterative way. In the paper, improvement of Q-learning method has been proposed using DE algorithm where guided randomness has been incorporated in the search space resulting fast convergence. Markov Decision Process (MDP), a mathematical framework has been used to model the problem in order to learn the large search space efficiently. The proposed algorithm exhibits better result in terms of speed and performance compare to basic Q-learning algorithm.

Abstract: learning concept has been incorporated by number of software and devices in the computer science and information industry. These software and devices are capable in decision making just like a human brain. This capability of decision making is govern by artificial intelligence techniques. These techniques follow many algorithms developed for decision making and machine learning. Decision making depends upon the profound training of contemporary data in a particular domain. Data plays a major and important part as one of the element in any machine learning algorithm. The main focus of this paper is on developing a machine learning algorithm that helps in training the available medical domain data to prepare a data model that negotiates with the query. This is achieved through the analysis of different machine learning methodologies like Support Vector Machines (SVM), Decision Trees and Recursive Partitioning (RP) algorithm and their model building processes. A new data mining and machine learning algorithm is proposed along with the performance analysis over the medical domain dataset. The analysis indicates that as the data size increases there is a continuous increase in algorithm accuracy but concurrently its time consumption also increases.

Abstract: It is difficult to apply traditional Q-learning algorithm to Multi-Agent environment, because in this case, the size of state-action space is so huge that it is hard to obtain the global optimal solution. In the paper, a task distribution based Q-learning algorithm is proposed to solve this problem. In this algorithm, at each learning step, it first distributes sub-task to each Agent dynamically. The Learning processes include the learning of task-distribution strategy and the learning of action-selection strategy synchronously, and every Agent shares the Q value table. Both Theoretical analysis and experimental results demonstrate that the proposed algorithm outperforms conventional Q-learning algorithm.