Basabi Chakraborty

Fuzzy Set Theoretic Measure for Automatic Feature Evaluation

The terms index of fuzziness, entropy, and ¿-ness, which give measures of fuzziness in a set, are used to define an index of feature evaluation in pattern recognition problems in terms of their intraclass and interclass measures. The index value decreases as the reliability of a feature in characterizing and discriminating different classes increases. The algorithm developed has been implemented in cases of vowel and plosive identification problem using formant frequencies and different S and ¿ membership functions.

Multiobjective route selection for car navigation system using genetic algorithm

Route planning is an important problem for a car navigation system. Given a set of origin-destination pair, there could be many possible routes for a driver. Search for shortest route from one point to another on a weighted graph is a well known problem and has several solutions like Dijkstra algorithm, Bellman-Ford algorithm etc. But in case of car navigation systems the shortest path may not be the best one from the point of view of drivers satisfaction. So, for a practical car navigation system in dynamical environment, we need to specify multiple and separate good (near optimal) choices according to multiple criteria which make the search space too large to find out the solution in real time by deterministic algorithms. Genetic algorithms (GA) are now widely used to solve search problems with applications in practical routing and optimization problems. GA includes a variety of quasi optimal solutions, which can be obtained in a given time. In this work we propose a GA based algorithm to find out simultaneously several alternate routes depending on different criterion according to drivers choice such as shortest path by distance, path which contains minimum number of turns, path passing through mountains or by the side of a river etc. The proposed algorithm has been evaluated by simulation experiment using real road map compared to other existing GA based algorithms. It has been found that the proposed algorithm is quite efficient in finding alternate non overlapping routes with different characteristics.

Feature subset selection by particle swarm optimization with fuzzy fitness function

Feature extraction or feature subset selection is an important preprocessing task for pattern recognition, data mining or machine learning application. Feature subset selection basically depends on selecting a criterion function for evaluation of the feature subset and a search strategy to find the best feature subset from a large number of feature subsets. Lots of techniques have been developed so far, mainly from statistical theory, still research is going on to find better solutions in terms of optimality and computational ease. Recently soft computing techniques are gaining popularity for solving real world problems for their more flexibility compared to statistical or mathematical techniques. In this work an algorithm based on particle swarm optimization with fuzzy fitness function has been proposed for getting optimal feature subset from a feature set with large number of features. Simple simulation experiments with two benchmark data sets show that the proposed method is similar in performance to the results reported earlier and is computationally less demanding in comparison to genetic algorithm, another population based evolutionary search technique proposed earlier for feature subset selection by author.

A new feature extraction technique for on-line recognition of handwritten alphanumeric characters

In this work a new method of feature extraction for an interactive and adaptive recognizer for on-line handwritten alphanumeric characters has been proposed. The system is suitable for use in conjunction with magnetic pen based devices for inputting data to a data processing system or a computer terminal. The features are extracted from dynamically changing locations of the writing device. The new method of feature extraction is simple, computationally light and fast enough for adaptive on-line use. Extracted features are robust with respect to all possible distortions like shape, size, and orientation. For simulation experiment, numerals 0-9 are used. A single hidden layer feed forward neural network trained by Quickprop algorithm, a variation of error back propagation is used for recognition. Very high recognition rates, even for highly distorted samples have been achieved confirming high generalization capability of the extracted feature set.

Feature Subset Selection by Neuro-rough Hybridization

Feature subset selection is of prime importance in pattern classification, machine learning and data mining applications. Though statistical techniques are well developed and mathematically sound, they are inappropriate for dealing real world cognitive problems containing imprecise and ambiguous information. Soft computing tools like artificial neural network, genetic algorithm fuzzy logic, rough set theory and their integration in developing hybrid algorithms for handling real life problems are recently found to be the most effective. In this worka neurorough hybrid algorithm has been proposed in which rough set concepts are used for finding an initial subset of efficient features followed by a neural stage to find out the ultimate best feature subset. The reduction of original feature set results in a smaller structure and quicker learning of the neural stage and as a whole the hybrid algorithm seems to provide better performance than any algorithm from individual paradigm as is evident from the simulation results.

A novel normalization technique for unsupervised learning in ANN

Unsupervised learning is used to categorize multidimensional data into a number of meaningful classes on the basis of the similarity or correlation between individual samples. In neural-network implementation of various unsupervised algorithms such as principal component analysis (PCA), competitive learning or self-organizing map (SOM), sample vectors are normalized to equal lengths so that similarity could be easily and efficiently obtained by their dot products. In general, sample vectors span the whole multidimensional feature space and existing normalization methods distort the intrinsic patterns present in the sample set. In this work, a novel method of normalization by mapping the samples to a new space of one more dimension has been proposed. The original distribution of the samples in the feature space is shown to be almost preserved in the transformed space. Simple rules are given to map from original space to the normalized space and vice versa.

Intraclass and interclass ambiguities (fuzziness) in feature evaluation

The terms Index of fuzziness, Entropy and @p-ness which give measures of fuzziness (ambiguity) in a set are used here to define an Index of Feature Evaluation in pattern recognition problems in terms of intraclass and interclass ambiguity. The index is seen to possess a lower value for the feature having more importance in characterising a class. The algorithm has been implemented on a speech recognition problem.

A review on application of data mining techniques to combat natural disasters

Abstract Thousands of human lives are lost every year around the globe, apart from significant damage on property, animal life, etc., due to natural disasters (e.g., earthquake, flood, tsunami, hurricane and other storms, landslides, cloudburst, heat wave, forest fire). In this paper, we focus on reviewing the application of data mining and analytical techniques designed so far for (i) prediction, (ii) detection, and (iii) development of appropriate disaster management strategy based on the collected data from disasters. A detailed description of availability of data from geological observatories (seismological, hydrological), satellites, remote sensing and newer sources like social networking sites as twitter is presented. An extensive and in-depth literature study on current techniques for disaster prediction, detection and management has been done and the results are summarized according to various types of disasters. Finally a framework for building a disaster management database for India hosted on open source Big Data platform like Hadoop in a phased manner has been proposed. The study has special focus on India which ranks among top five counties in terms of absolute number of the loss of human life.

A novel approach for estimation of optimal embedding parameters of nonlinear time series by structural learning of neural network

In this work a novel approach for estimation of embedding parameters for reconstruction of underlying dynamical system from the observed nonlinear time series by a feedforward neural network with structural learning is proposed. The proposed scheme of optimal estimation of embedding parameters can be viewed as a global non-uniform embedding. It has been found that the proposed method is more efficient for estimating embedding parameters for reconstruction of the attractor in the phase space than conventional uniform embedding methods. The simulation has been done with Henon series and three other real benchmark data sets. The simulation results for short term prediction of Henon Series and the bench mark time series with the estimated embedding parameters also show that the estimated parameters with proposed technique are better than the estimated parameters with the conventional method in terms of the prediction accuracy. The proposed technique seems to be an efficient candidate for prediction of future values of noisy real world time series.

Development of online handwriting recognition system: A case study with handwritten Bangla character

Developing efficient handwriting recognition systems that are fast and highly reliable is a challenging problem. This work represents the development of an online handwriting recognition system for Bangla script, widely used in eastern India and Bangladesh. In our approach, an online handwritten character/cluster is characterized by structure or shape based representation of a stroke in which a stroke is represented as a string of shape features. Using this string representation, an unknown stroke is identified by comparing it with a database of strokes using DTW (Dynamic Time Warping) technique. Identifying all the component strokes recognizes a full character. A recognition experiment has been conducted with a total of 495 classes on 20,873 data samples and 10 people as data contributors yielding 97.33% recognition rate with 2.18% misrecognition rate and 0.5% rejection rate.