Search

Data Mining is one of the greatest critical aspects of automatic disease diagnosis and disease prediction. It includes data mining algorithms and techniques to study medicinal data. Recently, liver disorders have frequently increased and liver illnesses have become one of the greatest deadly diseases in different countries. In this paper, we collected 534 instances of Iraqi liver patients from Baqubah Teaching Hospital. This collected dataset includes several elements such as Age, Total Bilirubin, Direct Bilirubin, Gender, Aspartate Aminotransferase, Total Proteins, and Albumin and Globulin ratio. This paper explores the early prediction of liver disease using various data mining algorithms. The main aim of this paper is to compute the performance of many data mining techniques and compare their performance with the performance of The Indian Liver Patient Dataset. We applied three classification algorithms and parameters like accuracy, precision, recall, and f-measure were investigated to evaluate the performance of these classifiers. [ABSTRACT FROM AUTHOR]

This paper focuses on finding the best classification algorithm model in the case study of student performance prediction and comparing the algorithm performance before and after using the normalization method. To achieve this goal, this study uses data mining classification techniques to analyse student performance at Vocational High School in 2020-2021. The steps of the research carried out include: data collection, data pre-processing, build algorithm models without using normalization and with using normalization, and final step are comparing algorithm performance before and after using normalization. The algorithms that will be used include: Random Forest, Decision Tree, Logistic Regression, SVM, Naive Bayes, and KNN. While the normalization methods used are Standard Scaler, Min-Max Scaler, and Robust Scaler. The result of this research is that the normalization method is able to significantly increase the accuracy of the model. Based on the tests and evaluations carried out, the normalization method using the Min-Max Scaler has the biggest impact in improving the overall model performance and the algorithm with the best performance is Random Forest. This paper reviews the effect of the normalization method to improve algorithm performance in predicting student performance, where based on previous research no one has used the normalization method to gain accuracy of the model which actually has a considerable impact on gaining accuracy of the model. [ABSTRACT FROM AUTHOR]

Much of the literature in knowledge discovery from data (KDD) focuses on algorithms that are faster and more accurate at capturing patterns in a given data set. However, answering a research question is fundamentally connected with how well the data is aligned with the questions being asked. Thus, data selection is one of the most important steps to ensure that models produced from the KDD process are useful in practice. A lack of documentation about the data selection rationale and the transformations needed to semantically align the data streams prevents others from reproducing the research and obfuscates development of best practices in data integration. Our goal in this paper is to provide KDD practitioners with a framework that brings together theories in provenance, information quality, and contextual reasoning, to enable researchers to achieve a semantically aligned dataset with data selection, description, and documentation based on an application‐focused assessment. [ABSTRACT FROM AUTHOR]

Data Mining and Analytics have transformed the way business is conducted. Prominent techniques of Data Mining include Clustering, Classification and Association Rule Mining. There are a lot of algorithms for cluster analysis. But all algorithms have some limitations especially for high dimensional dataset. In this paper modified the k-mean algorithm for better cluster accuracy and minimized the computational time processing the dataset for student academic performance. Students undergo immense stress and pressure to outscore their competitors. Higher and Technical Education has many challenges owing to voluminous data and various dimensions for acquisition and processing. Various Clustering Algorithms provide partition accuracy, computational time, lead time but lack visual assessment post partition. This paper proposes a variant of k-Means algorithm using comparative criteria of Partition accuracy (PA) parameter and computational time. [ABSTRACT FROM AUTHOR]

Automated diagnostic systems have become increasingly popular in the healthcare industry in recent years. In recent years, these systems have gained popularity for their diagnostic benefits. The operator-dependent existence of medical imaging systems can be minimized with this method, and the diagnostic process can be made more repeatable. In addition, it aids in the improvement of diagnostic accuracy. This effectively works with features that cannot be obtained by visual analysis or intuitive examinations (such as computational and statistical features). In this paper, we propose a new algorithm called "Optimized Weighted KNNI algorithm" (OWKNNI). In this research paper, we have addressed the topic of migration. This paper discusses the problem of dimensionality reduction and the use of feature selection algorithms, which is one of the most important tasks in feature engineering. The solution is given sorting out the problem is proposed as an algorithm. [ABSTRACT FROM AUTHOR]

The performance assessment of students is the most important work as this is directly related to the future of the students. This research paper focuses on structural study, analysis and identification of non academic characteristics and data mining techniques, specifically classification and clustering. These are required for designing a model which collectively gives performance assessment in the form of academic and non-academic characteristics. These characteristics need to be classified separately on the basis of different subcategories in each type of characteristics and data collected from the questionnaire from the stakeholders of this model students, teachers and parents. Identifying and using classification and clustering algorithms suitable to the structure of this assessment leads to required output of the research. [ABSTRACT FROM AUTHOR]

AI-based predictive data mining algorithms play an important role in the prediction-role of the two-faceted roles of data mining (classification as the other role). Having a rigid predictive algorithm is tantamount to a successful and near-human predictive decision in knowledge discovery among databases being implemented among government, commercial and educational sectors. In this paper, three (3) of the most commonly used AI-based predictive data mining algorithms, namely: instance-based (nearest neighbor), statistical (naive Bayes), and Bayesian networks were evaluated using academic admission data in order to assess which predictive algorithm is best suited for predicting an ideal collegiate course for a student based on the student's high school grades and historical admission records. This study sought also to find out which among the four predictive algorithms can be used in terms of volume of historical data and number of data attributes. [ABSTRACT FROM AUTHOR]

Numerous researchers have applied nature-inspired population-based metaheuristics for solving optimization problems including data clustering. However, the issues of premature convergence and slow convergence rate can still occur when these promising search methods are applied to complex and large data-clustering problems, including the evaporation-rate based water cycle algorithm, WCAER. In this paper, a recently proposed hybrid version of WCAER in conjunction with a local search method named Hookes and Jeeves method was further tested to perform data clustering for large dataset. The proposed hybrid algorithm is experimented on the gas turbine emission data that contains 36733 instances of 11 sensor measures aggregated over one hour, from a gas turbine in Turkey, available from the UCI machine-learning repository. The simulation results confirm the superiority of the hybrid method as an efficient and reliable algorithm to solve gas turbine clustering problem, in comparison to the original evaporation-rate based water cycle algorithm, in terms of solution quality as well as computational performance formulated from two applied objective functions namely the Euclidean distance and the Davies-Bouldin index. Thus, the outcome of the study generally provides some performance evaluation of WCAER for large dataset when applied in cluster analysis, a valuable data analysis and data mining technique. It is hoped that other metaheuristics could be applied to the 5 years span large dataset for performance comparison with other further considerations such as using various suitable objective functions and cluster validity indices. [ABSTRACT FROM AUTHOR]

The improving of a company's earnings is considered as one of the important Challenges that facing these companies in terms of the amounts of their manufactured products. Therefore, it is very essential to determine which units of production that gives the reasonable profit. In this paper, data Mining (DM) and decision tree were used to define the products quantities of the Iraqi carpet company in the year 2016 that led to the acceptable profit using decision tree algorithm. This said algorithm is determined the product amounts that can gave the required profit namely; low permeable carpet, high embroidered blanket and high woven blanket with their quantities138, 1172, and 5025 respectively. In order to check the validity of the acceptable profit, these products were tested by the Naive Bayes algorithm which check their probabilities as probability of P (yes) and P (no) got 0.0208 and 0 respectively. [ABSTRACT FROM AUTHOR]

Typically, refactoring are expressed in high-level, general purpose programming languages, which are impractical to verify because they are too complex to have tractable formal semantics definitions. Refactoring can be defined in various formats at various abstraction levels, the sorting algorithm is the most important and powerful in most or all programs language and database, the mighty importance of the sorting towards data mining that needs ordered the datasets elements to be fast particularly in preprocessing data. Methods : The challenges side incur of this paper summed up development the sort algorithm many times, using different refactoring ways as sequentially method sometimes with list comprehension moreover recursive listing, further with parallel computing, Implementation and execute different methods for the sorting algorithm code on several different computers with different in CPU to achieve the best rate of pure results for implementation under the Erlang code. Proposed solution : The purpose of this paper is to conduct a comparison between methods by evaluate the runtime execution prepared code that writing under Erlang functional programming, the research effort comparative analysis between the approaches adopted many member of the sort algorithm and proposed a new technique that works ordered the list by minimum elements with recursive the list comprehensive. Result: then this paper will a prove the Quick Sort technique is the fastest compared with the others. [ABSTRACT FROM AUTHOR]

Self Organization Map (SOM) is an automatic tool in data analysis in data mining,it is used to explore the multi-dimentional data which simplifies complexity and produce meaningful relation with each other or high dimentional into low dimentional .The powerful method of SOM i.e learning method results excellent performance .the SOM algorithum have various steps from starting stage to the final neuron and their weight updation and modification ,these procedure resultant a lot of complexity according to the parameters on the basis of experiments, this paper will compare and discuss various parameters and their result or factors that can improve and refine the image through various process of SOM. [ABSTRACT FROM AUTHOR]

This paper provides a metaheuristic-based of classification rules. Indeed, The classification problem is a data mining task which belongs to the NP-Complete problem class. This is a combinatorial problem which may find good solutions using metaheuristics approaches. We are interested in adapting the Bee Swarm Optimisation (BSO) algorithm which is relatively a new metaheuristic. Our aim is to discover a classification model. This BSO approach already gave some interesting solutions for the well-known SAT problem and we will show the efficiency of the method for the Problem of the Extraction of Classification Rules also. [ABSTRACT FROM AUTHOR]

This paper uses the matrix compression algorithm and weight analysis algorithm for reference and proposes an improved matrix pruning and weight analysis Apriori algorithm. After the transactional database is scanned for only once, the algorithm will construct the boolean transaction matrix. Through the calculation of one figure in the rows and columns of the matrix, the infrequent item set is pruned, and a new candidate item set is formed. Then, the item’s weight and the transaction’s weight as well as the weight support for items are calculated, thus the frequent item sets are gained. The experimental result shows that the improved Apriori algorithm not only reduces the number of repeated scans of the database, but also improves the efficiency of data correlation mining. [ABSTRACT FROM AUTHOR]

The classical method of finding associations in the data matrix with the given support and reliability may not lead to finding all relevant facts in the data. This paper deals with a modified method of searching for so called aggregated associations, which can also find other non-trivial facts. The paper describes an algorithm that finds hypotheses of the form of aggregated associations and demonstrates its use in a practical example. [ABSTRACT FROM AUTHOR]

The Data mining algorithms have become vital to researchers in science, engineering, medicine, business, search and security domains. In recent years, there has been a terrific raise in the size of the data being collected and analyzed. Classification is the main difficulty faced in data mining. In a number of the solutions developed for this problem, most accepted one is Decision Tree Classification (DTC) that gives high precision while handling very large amount of data. This paper presents VLSI implementation of flexible architecture for Decision Tree classification in data mining using c4.5 algorithm. [ABSTRACT FROM AUTHOR]

In order to extract, transform and load large scale data from heterogeneous data sources into data warehouse efficiently, the SETL system is designed and implemented in this paper. By using PERL subroutine attribute and data partition, SETL can implement ETL job easily and perform ETL job efficiently, and the plug-in design makes SETL with high scalability, and the design that performing one ETL job in one ETL pipeline makes SETL with distribution environment support. For illustration, one ETL job example is utilized to show the scalability in designing ETL job and show the high efficiency in processing large scale data. Experiments prove that SETL can extract, transform, and load large scale data into data warehouse efficiently. The SETL system simplified the ETL job design and implementation and can deal with heterogeneous data sources flexibly. It is a light-weighted, scalable and high-performance ETL system. [ABSTRACT FROM PUBLISHER]

This paper analyzes the current situation of the intrusion detection system, which is the basis to put forward that data mining technology is to be applied to the intrusion detection system in terms of the problems of the traditional intrusion detection system. Meanwhile, the paper designs the intrusion detection model of data mining. With the study on intrusion detection and data mining, the algorithm of classical relation with clustering in the characteristics on intrusion detection system is improved and optimized. [ABSTRACT FROM AUTHOR]

Clonal Selection Algorithm (CSA) is a sturdy evolutionary algorithm that mimics the biological immune system mechanism, frequently implemented as a searching technique in solving complex modelling tasks such as pattern recognition and combinatorial optimization. Subsequently, 3-Satisfiability logic mining is the fundamental medium to induce real data sets in a more precise manner. In this paper, a modified clonal selection algorithm with Hopfield neural network (HNN) is proposed for 3-Satisfiability data mining. Somatic hypermutation operator in CSA can reduce the iterations during HNN training phase and retrieve the best logical rule for a minimal testing error. In addition, 3- Satisfiability logical rule provides better attributes representation of the data set. The proposed method will be applied on Amazon's Employees Resources Access data set to predict the approval or denial for an unseen set of employees in the future. The performance evaluation of the proposed method will be compared with exhaustive search (ES) by computing plausible errors such as root mean square error (RMSE), mean absolute error (MAE), sum of squared error (SSE), accuracy and computational time. The computational simulations are carried out by manipulating different number of neurons to verify the capability of the proposed method in data mining. The experimental results have shown a better performance of the proposed model in training and retrieval phase of the Amazon's Employees Resources Access data set in comparison with ES used in this research. [ABSTRACT FROM AUTHOR]

This paper addresses smart data mining techniques for analyzing the processing of selected powder coatings on co-rotating twin screw extruders. The influences of different machinery and process parameters on optical quality criteria such as degree of gloss and color are analyzed. Experiments were performed on laboratory and production scale co-rotating twin screw extruders using different screw configurations. Decision trees are developed in order to analyze the process and to determine the essential influencing factors on the formation of the degree of gloss. For further analysis of the process and to enable optimal up-scale from laboratory to production scale analytic equations are evolved by using symbolic regression based on heuristic optimization algorithms. [ABSTRACT FROM AUTHOR]

For the issues of the rare itemsets mining in big data, this paper proposed a rare itemsets mining algorithm based on RP-Tree and Spark framework. Firstly, it arranged the data vertically according to the transaction identifier, in order to solve the defects of scan the entire data set, the vertical datasets are divided into frequent vertical datasets and rare vertical datasets. Then, it adopted the RP-Tree algorithm to construct the frequent pattern tree that contains rare items and generate rare 1-itemsets. After that, it calculated the support of the itemsets by scanning the two vertical data sets, finally, it used the iterative process to generate rare itemsets. The experimental show that the algorithm can effectively excavate rare itemsets and have great superiority in execution time. [ABSTRACT FROM AUTHOR]

As a key technique in data mining, classification algorithm was received extensive attention. Through an experiment of classification algorithm in UCI data set, we gave a comparison analysis method for the different algorithms and the statistical test was used here. Than that, an adaptive diagnosis model for preventive electricity stealing and leakage was given as a specific case in the paper. [ABSTRACT FROM AUTHOR]

The developed multidimensional data mining algorithm integrates logical and statistical computing. It provides interactive and exploratory environment for mining spatial and temporal features of time series datasets. The algorithm is applied on Crop production database which stores spatial time series on crop production in EU-Balkan states. The paper discusses the mined information on crop production quantity and area harvested in Bulgaria at national and regional level. The data mining algorithm and software can be used to analyze spatial time series when studying the dynamics of social systems, economical activities and natural resources exploitation. [ABSTRACT FROM AUTHOR]

One of the challenges of research on Technology for Human Learning (THL) and the competency based pedagogy is the personalization of learning. This customization begins with the design of educational systems, applying the strategies defined to monitor and assist the learner in his learning by evaluating his knowledge, skills and detecting and analyzing errors. In this respect, formative evaluation is the process used to capture data on the strengths and weaknesses of a learner during a learning period. These data, to be useful, must be objectively analyzed so that they can be used to manage the following sessions. Data mining is a field at the intersection of statistics and information technology (databases, artificial intelligence, learning etc) whose goal is to discover "interesting" knowledge structures from a set of data. Following our research work on evaluation process and THL, applied to e-learning systems [2] [3] [4], we discuss in this paper a learning cycle with feedback loop integrating formative evaluation followed by the application of data mining algorithms after each learning session. We experimented with a set of tests, the exploration of learners' errors, obtained from a self e-learning by doing tool for the algorithmic domain. We particularly present, in this work, the results given by the application of the C4.5 algorithm for classification and the A Priori one for association rules deduction. These two algorithms were executed via the tool Weka after a data preparing process integrating a relational algebra calculation and format adjustment actions to get adequate input data for Weka algorithms. The results given by these experiments have proved interesting knowledge to complete the learner's profile and facilitate decisions about further learning / teaching / tutoring particularly in the case of distance education. [ABSTRACT FROM AUTHOR]

Data mining is the discipline of systematically reviewing datasets to determine what patterns, concurrences and/or rule sets can be discovered. In an effort to understand the effectiveness of a proposed mining methodology a well defined dataset is required for model verification. To this end, a fully-controlled simulation was created by which several different feature selection algorithms were evaluated. In this paper, we present a comprehensive comparison between attribute selection methods when noise, missing values and multicollinearity are in question. Our results show that "Relief" and "information gain" have outperformed other feature selection methods available in Weka when considering both sensitivity and specificity measures. We have evaluated the following features selection methods: J48, Relief, information gain, consistency based feature selection and correlation based feature selection to see which one handles additive noise better. The sensitivity of consistency based feature selection was 11% higher than the average sensitivity of other methods. However, it's specificity was 37% lower than that of the average. It is important to note that sensitivity or specificity alone does not give enough support to a method to say that it is the best way to handle the data. The best method, when both sensitivity and specificity are considered, was information gain. This method outperformed the average of other methods by 1% and 20.5% when we considered its sensitivity and specificity, respectively. Also in this regard, a goal of our study was to see which feature selection method outperforms within the missing set of values. In this case, when looking again at sensitivity and specificities, Relief and information gain proved to outperform the other methods of our study by 7.2% and 12.4%, respectively. Our studies also show that when multicollinearity is embedded into the fully controlled dataset without any noise and missing values, the correlation based feature selection outperforms other methods. In summary Relief and information gain performed the best in all three situations in terms of its sensitivity and specificities. [ABSTRACT FROM AUTHOR]

Clustering has played a very important role in data mining. In this paper, a fast incremental clustering algorithm is proposed by changing the radius threshold value dynamically. The algorithm restricts the number of the final clusters and reads the original dataset only once. At the same time an inter-cluster dissimilarity measure taking into account the frequency information of the attribute values is introduced. It can be used for the categorical data. The experimental results on the mushroom dataset show that the proposed algorithm is feasible and effective. It can be used for the large-scale data set. [ABSTRACT FROM AUTHOR]

As a simple classification method VSM has been widely applied in text information processing field. There are some problems for traditional VSM to select a refined vector model representation, which can make a good tradeoff between complexity and performance, especially for incremental text mining. To solve these problems, in this paper, several improvements, such as VSM based on improved TF, TFIDF and BM25, are discussed. And then maximum mutual information feature selection is introduced to achieve a low dimension VSM with less complexity, and at the same time keep an acceptable precision. The experimental results of spam filtering and short messages classification shows that the algorithm can achieve higher precision than existing algorithms under same conditions. [ABSTRACT FROM AUTHOR]

Cardiovascular diseases are the leading causes of death worldwide, from which heart attack and coronary artery disease are the most common ones. Both diseases are direct causes of Heart Failure, which affects the heart's pumping job, leading to an abnormal distribution of nutriments and oxygen to the body. Data mining techniques have been used to predict different heart diseases. This paper describes the application of a data mining technique, namely the sequence prediction algorithm, CPT+ or Compact Prediction Tree plus, to predict the absence and the presence of heart failure in patients. To evaluate the efficiency of adopted technique, we ran the algorithm in a dataset from the UCI Machine Learning Repository, containing a list of 14 attributes, including the attribute to be predicted. The evaluation showed promising results in term of predicting the absence and presence of heart disease leading to heart failure. [ABSTRACT FROM AUTHOR]

Research Question There has been explosive growth in the amount of data and information available to us and about us ("big data"). Our daily activities produce vast quantities of data that provide raw material for companies to create detailed profiles about us. Companies benefit from fewer wasted marketing dollars and greater capability to make inferences about our behavior. People benefit because they receive personalized offerings and access to massive amounts of information, free of charge (a goldmine). The accumulation of data is accelerating and the types of data are varied and vast. Along with the explosion of data from company efforts, there also is an explosion of data from what people share about themselves. Although people share, they also report concerns about disclosure of personal data (privacy paradox). Thus, while there are benefits from the profusion of personal data, there are potential costs related to how people's personal data and information are used (a minefield). This paper seeks to "spark conversation" by examining recent trends in data collection and aggregation and to explore potential harms that give rise to privacy concerns and to identify costs and tradeoffs people are willing to accept to offset potential harms or to gain specific benefits. Summary of Findings Kokolakis (2015) notes that while the privacy paradox is no longer a paradox, it still remains a complex phenomenon that is not fully explained. Complicating efforts to explain this paradox are recent developments in data collection and aggregation that have become more sophisticated and less transparent (e.g., facial recognition, cognitive computing, smart algorithms). With these developments, the ability to track and monitor people without their awareness is increasing, which raises the possibility of nudging or controlling people's behaviors. This is both promising and problematic. Promising in that we see the design of personalized product offerings that anticipate our needs (e.g., a smarter Siri, Hello Barbie) and that nudge us to engage in positive behaviors (e.g., exercise). Further, smart algorithms rely on "scientific" data rather than biases or heuristics and thus eliminate the need to use social constructs to profile people and predict behavior (although some evidence may suggest otherwise). While nudging clearly can lead to positive outcomes, it may be problematic in that these same smart tools may limit our choices (e.g., high interest credit cards or mortgages). This, too, is personalization, although not at its best. Key Contributions As the field of big data changes, new developments will continue to unfold. Many of the developments reviewed in this paper rely on methods of data collection and aggregation that make it more difficult for people to balance the loss of privacy with increased personalization. This paper seeks to explore this complex phenomenon further. One of the author's starting points will be Solove's (2006) taxonomy of privacy that provides a framework for examining the range of problems that can arise from the "activities that technology enables." While his framework was designed to help understand privacy problems and protections from a legal perspective, his taxonomy provides a useful framework for examining the big data as goldmine versus minefield dichotomy. Unfettered access to people's data allows companies to build "actionable intelligence" (a goldmine); but this same access makes it more difficult for people to assess the costs of the use of their data (a minefield). Use of big data works well when people's interests are aligned with those of the data collector. However, if not aligned, this raises the question of whose interests are served, particularly if people are unaware of the data collection. This paper seeks to better understand this relationship. [ABSTRACT FROM AUTHOR]

Cluster analysis is an important category of data mining issues. Clustering techniques are frequently applied also in business and economy, for example in customer's classification, enterprises grouping or in other economic data analyses. The paper submits a new clustering method, which we have named Firefly Clustering Algorithm (FCA), based on a recently developed and commonly applied optimisation technique - Firefly Algorithm (FA) that simulates natural behaviour of living fireflies. After brief presentation of the original idea of classical FA as an optimisat ion technique, we submit the description of FCA - the proposed clustering method. The algorithm has been implemented in form of our computer program written in C++. Next we show a financial utilisation of FCA as a tool for cluster analysis of selected stock companies quoted on Warsaw Stock Exchange, with taking into account two factors (as input variables): return rate and risk. Results have been visualised and discussed. Directions of further researches have been outlined. Although here we have considered a 2-dimensional problem (i.e. a dataset with only 2 numerical features), the presented FCA method is universal and can be applied for any number of inputs (obviously with limited visualisation capabilities). [ABSTRACT FROM AUTHOR]

Recently, medical data mining has become one of the most popular topics in the data mining community. This is due to the societal importance of the field and also the particular computational challenges posed in this domain of data mining. Early researches concentrated on sequential heuristics and later moved to meta-heuristic approaches due to the ability of these approaches to generate better solutions. The aim of this paper is to introduce the basic principles of a new meta-heuristic algorithm called Electromagnetism-like Mechanism (EMag) for neural network training. EMag simulates the electromagnetism theory of physics by considering each data sample to be an electrical charge. For neural network, EMag simulates the attraction-repulsion mechanism of each weight connection as charge partials to move towards the optimum without being trapped into local minimum. The performance of the proposed algorithm is evaluated in 12 of benchmark classification problems, and the computational results show that the proposed algorithm performs better than the standard back propagation algorithm. [ABSTRACT FROM AUTHOR]

In this paper we design a system that adopts a novel approach for emotional classification from human dialogue based on text and speech context. Our main objective is to boost the accuracy of speech emotional classification by accounting for the features extracted from the spoken text. The proposed system concatenates text and speech features and feeds them as one input to the classifier. The work builds on past research on music mood classification based on the combination of lyrics and audio features. The innovation in our approach is in the specific application of text and speech fusion for emotion classification and in the choice of features, Furthermore, in the absence of benchmark data, a dataset of movie quotes was developed for testing of emotional classification and future benchmarking. The comparison of the results obtained in each case shows that the hybrid text-speech approach achieves better accuracy than speech or text mining alone. [ABSTRACT FROM PUBLISHER]

The present study aims at investigating the different Data mining learning models for different medical data sets and to give practical guidelines to select the most appropriate algorithm for a specific medical data set. In practical situations, it is absolutely necessary to take decisions with regard to the appropriate models and parameters for diagnosis and prediction problems. Learning models and algorithms are widely implemented for rule extraction and the prediction of system behavior. In this paper, some of the well-known Machine Learning(ML) systems are investigated for different methods and are tested on five medical data sets. The practical criteria for evaluating different learning models are presented and the potential benefits of the proposed methodology for diagnosis and learning are suggested. [ABSTRACT FROM AUTHOR]

Among difficulties encountered by modern shopping recommenders one of the most important is the long tail shape of sold items which is also related to cold-start issues. Various approaches including content-based recommendations attempt to overcome this problem which has a serious impact on the accuracy of recommendations especially when new products are continuously added to the catalogue. A hybrid approach that combines recommendations based on highly correlated pairs of concrete items (existing items) with abstract items (based on the item taxonomy) is proposed in this paper (Abstract-Concrete Recommendation-ACReco). The ACReco algorithm is evaluated on an in-house and a benchmark of 0-1 Data (the Ta Feng data). "Given-n" protocols experimental results show significant improvements in both the recommendation accuracy and the recommendation of products in the long tail. The overall efficiency of the approach relies on the available Product Taxonomy. [ABSTRACT FROM AUTHOR]

This paper investigated the performance of four of the top rated classification techniques, namely, Naive Bayes, K-Nearest Neighbour (K-NN), Decision Tree (ID3) and CART as stated and voted for in the Top 10 data mining algorithms in [20]. We analysed their performance on 10 real nominal datasets and discovered their strengths and weaknesses based on different features of the datasets. It could be seen using the produced results that different features of datasets would determine how fast and accurate the different classification techniques are. This study and the recommendations we provide at the end would help data mining practitioners on deciding the appropriate technique based on the dataset's number of attributes, instances, class labels and the total size. [ABSTRACT FROM AUTHOR]

Currently, the implementation of WEB caching is mostly based on traditional cache updating algorithms. However, due to the diversity of the WEB traffic pattern, the traditional algorithm s for cache updating can not be used in WEB environment effectively. Literature (F. Bonchi, 2001) provides a Web updating algorithms based on Decision Trees, in this paper, we provide a web policy, based on Genetically Evolved Decision Trees, the result of experiment indicates this policy improves the updating efficiency of the Web pages. [ABSTRACT FROM AUTHOR]

Due to the magnitude and complexity of design and manufacturing processes, it is unrealistic to expect that models and simulations can predict all aspects of silicon behavior accurately. When unexpected behavior is observed in the post-silicon stage, one desires to identify the causes and consequently identify the fixes. This paper studies one formulation of the design-silicon mismatch problem. To analyze unexpected behavior, silicon behavior is partitioned into two classes, one class containing instances of unexpected behavior and the other with rest of the population. Classification rule learning is applied to extract rules to explain why certain class of behavior occurs. We present a rule learning algorithm that analyzes test measurement data in terms of design features to generate rules, and conduct controlled experiments to demonstrate the effectiveness of the proposed approach. Results show that the proposed learning approach can effectively uncover rules responsible for the designsilicon mismatch even when significant noises are associated with both the measurement data and the class partitioning results for capturing the unexpected behavior. [ABSTRACT FROM AUTHOR]

One of the most important data mining problems is mining association rules. In this paper, we consider discovering association rules from large transactional databases. The problem of discovering association rules can be decomposed into two subproblems: find large itemsets and generate association rules from large itemsets. The second subproblem is easier one and the complexity of discovering association rules is determined by complexity of discovering large itemsets. In this paper, we suggest improvements of the Apriori algorithm, which is one of the most famous algorithms for discovering large itemsets. Actually, we suggest the original procedure for large itemsets generation, which produce less number of candidate itemsets and which is efficient than the appropriate procedure of the Apriori algorithm. For its implementation, we suggest modified sort-merge-join algorithm. Also, we propose a way how to algorithm finishes in tree database scans. [ABSTRACT FROM AUTHOR]

Community mining in heterogeneous social networks is different from traditional community mining approaches in that it assumes the existence of more than one kind of relations between objects. It consists of two steps. Relation extraction in which the goal is to combine existing relations to form a relation based on the query provided by the user; and community mining which identifies community structure in the extracted relation. In this paper we are mainly concerned with the first step. Previously proposed algorithms for relation extraction are either not applicable to large data sets or can handle simple queries only. In this paper, we propose two genetic algorithms which use different kind of operators to search the problem space. These algorithms work in general situations, and are more scalable, interpretable, and extensible than previous methods. Experiments and analysis of the results on DBLP dataset show the power of our algorithms in handling complex queries in large data sets. [ABSTRACT FROM AUTHOR]

This paper is on market forecast summarization; in particular, it focuses on the extraction and comprehensive visualization of market forecasts from Internet data. Market forecast extraction is essential for investment strategists who base their investment decisions on predicted market developments. Against this background, we break down the complex analysis of market developments into three specific questions related to turnover estimation. The contributions of our paper relate to automated document time analysis, information quality assessment, and a new time extraction algorithm. Moreover, we will present a prototypic system that operationalizes our ideas. The paper concludes with a case study from the field of new technology assessment, in order to demonstrate functions of our prototype. [ABSTRACT FROM AUTHOR]

Within strategic foresight literature, customer foresight still shows a low capability level. In practice, especially in b2b industries analyzing an entire customer base in terms of future customer potential is often done manually. Therefore, we present a single case study based on a quantitative customer foresight project conducted by a manufacturing company. Along a common data mining process, we highlight the application of machine learning algorithms on an entire customer data base which consists of customer and product related data. The overall benefit of our research is threefold. Major result is a prioritization of 2,300 worldwide customers according to their predicted technical affinity and suitability for a new machine control sensor. Furthermore, we explain necessary requirements and steps for realizing a customer foresight project. Finally, we provide a detailed catalogue of measures on how to strategically deal with the identified high potentials and all related customer insights. [ABSTRACT FROM AUTHOR]

Within the large body of biomedical knowledge, recent findings and discoveries are most often presented as research articles. Their number has been increasing sharply since the turn of the century, presenting ever-growing challenges for search and discovery of knowledge and information related to specific topics of interest, even with the help of advanced online search tools. This is especially true when the goal of a search is to find or discover key relations between important concepts or topic words. We have developed an innovative method for extracting key relations between concepts from abstracts of articles. The method focuses on relations between keywords or topic words in the articles. Early experiments with the method on PubMed publications have shown promising results in searching and discovering keywords and their relationships that are strongly related to the main topic of an article. [ABSTRACT FROM AUTHOR]

In this paper we propose a complement to the Rep-Index reputation metric for researchers, using data mining techniques specific for selecting a subset of the most relevant elements to Rep-Index. We redid the experiments with the original author's data and found the best algorithm. At the end we propose the new weights for classification of researchers with correct classification rate higher than the original Rep-Index. For Computer Science the best selection attribute is InfoGain. The Spearman's correlation is 0.5520 against 0.3932 for Rep-Index with original weights. The results show that Rep-Index is adaptable, flexible and the approach proposal complements the original index to offer better classification for researchers. [ABSTRACT FROM AUTHOR]

The means of data mining and machine learning tasks are important topics in signal processing fundamentals. An example of such tasks is motif discovery. This paper presents an efficient method for shift-invariant feature extraction in time-series motif discovery. The proposed method initiates from the machine learning procedure and tackles the drawbacks of existing methods. Moreover, the efficacy of the novel approach is benchmarked against various algorithms and data from diverse fields. [ABSTRACT FROM AUTHOR]