P. S. V. S. Sai Prasad

Scalable Quick Reduct Algorithm: Iterative MapReduce Approach

Feature selection by reduct computation is the key technique for knowledge acquistion using rough set theory. Existing MapReduce based reduct algorithms use Hadoop Map Reduce framework, which is not suitable for iterative algorithms. Paper aims to design and implementation of Iterative MapReduce based Quick reduct algorithm using Twister framework. The proposed In_MRQRA Algorithm has partial granular level computations at mappers and granular computations at reducer. Experimental analysis on KDD-Cup99 dataset empirically established the relevence of proposed approach.

Scalable IQRA_IG Algorithm: An Iterative MapReduce Approach for Reduct Computation

Feature Selection is an important preprocessing step in any machine learning model construction. Rough Set based feature selection (Reduct) methods provide efficient selection of attributes for the model without loss of information. Quick Reduct Algorithm is a key Reduct computation approach in Complete Symbolic Decision Systems. Authors have earlier implemented a scalable approach for Quick Reduct Algorithm as In-place MapReduce based Quick Reduct Algorithm using Twister’s Iterative MapReduce Framework. Improved Quick Reduct Algorithm is a standalone extension to Quick Reduct Algorithm by incorporating Trivial Ambiguity Resolution and Positive Region Removal. This work develops design and implementation of distributed/parallel algorithm for Improved Quick Reduct Algorithm by incorporation of Trivial Ambiguity Resolution and Positive Region Removal in In-place MapReduce based Quick Reduct Algorithm. Experiments conducted on large benchmark decision systems have empirically established the significance of computational gain and scalability of proposed algorithm in comparison to earlier approaches in literature.

An Efficient Gaussian Kernel Based Fuzzy-Rough Set Approach for Feature Selection

Fuzzy-rough set based feature selection is highly useful for reducing data dimensionality of a hybrid decision system, but the reduct computation is computationally expensive. Gaussian kernel based fuzzy rough sets merges kernel method to fuzzy-rough sets for efficient feature selection. This works aims at improving the computational performance of existing reduct computation approach in Gaussian kernel based fuzzy rough sets by incorporation of vectorized (matrix, sub-matrix) operations. The proposed approach was extensively compared by experimentation with the existing approach and also with a fuzzy rough set based reduct approaches available in Rough set R package. Results establish the relevance of proposed modifications.

An Efficient Approach for Fuzzy Decision Reduct Computation

Fuzzy rough sets is an extension of classical rough sets for feature selection in hybrid decision systems. However, reduct computation using the fuzzy rough set model is computationally expensive. A modified quick reduct algorithm (MQRA) was proposed in literature for computing fuzzy decision reduct using Radzikowska-Kerry fuzzy rough set model. In this paper, we develop a simplified computational model for discovering positive region in Radzikowska-Kerry’s fuzzy rough set model. Theory is developed for validation of omission of absolute positive region objects without affecting the subsequent inferences. The developed theory is incorporated in MQRA resulting in algorithm Improved MQRA (IMQRA). The computations involved in IMQRA are modeled as vector operations for obtaining further optimizations at implementation level. The effectiveness of algorithm(s) is empirically demonstrated by comparative analysis with several existing reduct approaches for hybrid decision systems using fuzzy rough sets.

Seed based fuzzy decision reduct for hybrid decision systems

Fuzzy rough sets is an extension to classical rough sets. The fuzzy rough set model is useful in feature selection for hybrid decision systems. Fuzzy decision reduct uses Radzikowskas Fuzzy Rough Set model for feature selection in hybrid decision systems. The computational complexity of fuzzy decision reduct computation makes it not suitable for large hybrid decision systems. In this paper, an approach is developed for computing fuzzy decision reduct by seed reduct using a suitable discretization of quantitative conditional attributes. Fuzzy decision reduct is computed for original decision system by evolving over seed reduct. Theoretical analysis and experimental results on benchmark decision systems validate that the method has achieved significant computational gains over normal approach without loss of classification accuracy.

Impact analysis of Jensen and Sk pal fuzzification in classification

In this paper, we compare fuzzy-rough and Ant fuzzy-rough feature selection algorithms with respect to reduct and corresponding classification accuracy. It is known that, prior to applying these fuzzy rough feature selection algorithms; fuzzification of the dataset has to be done. For which, we have adopted Jensens fuzzification method (used by him for his work) and Sk pal fuzzification method to fuzzify the data. The comparison is done specifically with respect to these fuzzification methods on benchmark datasets. In addition, it shows that Sk pals fuzzification gives better results than Jensen fuzzification method.

Hashing Supported Iterative MapReduce Based Scalable SBE Reduct Computation

Feature Selection plays a major role in preprocessing stage of Data mining and helps in model construction by recognizing relevant features. Rough Sets has emerged in recent years as an important paradigm for feature selection i.e. finding Reduct of conditional attributes in given data set. Two control strategies for Reduct Computation are Sequential Forward Selection (SFS), Sequential Backward Elimination(SBE). With the objective of scalable feature seletion, several MapReduce based approaches were proposed in literature. All these approaches are SFS based and results in super set of reduct i.e. with redundant attributes. Even though SBE approaches results in exact Reduct, it requires lot of data movement in shuffle and sort phase of MapReduce. To overcome this problem and to optimize the network bandwidth utilization, a novel hashing supported SBE Reduct algorithm(MRSBER_Hash) is proposed in this work and implemented using Iterative MapReduce framework of Apache Spark. Experiments conducted on large benchmark decision systems have empirically established the relevance of proposed approach for decision systems with large cardinality of conditional attributes.

Third Order Backward Elimination Approach for Fuzzy-Rough Set Based Feature Selection

Two important control strategies for Rough Set based reduct computation are Sequential Forward Selection (SFS), and Sequential Backward Elimination (SBE). SBE methods have an inherent advantage of resulting in reduct whereas SFS approaches usually result in superset of reduct. The fuzzy rough sets is an extension of rough sets used for reduct computation in Hybrid Decision Systems. The SBE based fuzzy rough reduct computation has not attempted till date by researchers due to the fuzzy similarity relation of a set of attributes will not typically lead to fuzzy similarity relation of the subset of attributes. This paper proposes a novel SBE approach based on Gaussian Kernel-based fuzzy rough set reduct computation. The complexity of the proposed approach is the order of three while existing are fourth order. Empirical experiment conducted on standard benchmark datasets established the relevance of the proposed approach.

Scalable MapReduce-based Fuzzy Min-Max Neural Network for Pattern Classification

Fuzzy Min-Max Neural Network (FMNN) is a pattern classification algorithm which incorporates fuzzy sets and neural network. It is most suitable for online algorithms. Based on this, a MapReduce-based Fuzzy Min-Max Neural Network (MRFMNN) algorithm for pattern classification is proposed using Twister framework. MapReduce approach is used for scaling up the FMNN for massive large scale datasets. We used standard membership, expansion and the contraction functions of the traditional FMNN algorithm. The performance of the MRFMNN is tested by using several benchmark and synthetic datasets against the traditional FMNN. Results empirically established that MRFMNN achieves significant computational gains over FMNN without compromising classification accuracy.

A New Preprocessor to Fuzzy c-Means Algorithm

The fuzzy clustering scenario resulting from Fuzzy c-Means Algorithm (FCM) is highly sensitive to input parameters, number of clusters c and randomly initialized fuzzy membership matrix. Traditionally, the optimal fuzzy clustering scenario is arrived by exhaustive, repetitive invocations of FCM for different c values. In this paper, a new preprocessor based on Simplified Fuzzy Min-Max Neural Network (FMMNN) is proposed for FCM. The new preprocessor results in an estimation of the number of clusters in a given dataset and an initialization for fuzzy membership matrix. FCM algorithm with embeded preprocessor is named as FCMPre. Comparative experimental results of FCMPre with FCM based on benchmark datasets, empirically established that FCMPre discovers optimal (or near optimal) fuzzy clustering scenarios without exhaustive invocations of FCM along with obtaining significant computational gains.