We investigate techniques to combine multiple representations of a handwritten digit to increase classification accuracy without significantly increasing system complexity or recognition time. We compare multiexpert and multistage combination techniques and discuss in detail in a comparative manner methods for combining multiple learners: Voting, mixture of experts, stacking, boosting and cascading. In pen-based handwritten character recognition, the input is the dynamic movement of the pentip over the pressure sensitive tablet. There is also the image formed as a result of this movement. On a real-world database, we notice that the two multi-layer perceptron (MLP) neural network-based classifiers using separately these representations make errors on different patterns implying that a suitable combination of the two would lead to higher accuracy. Thus we implement and compare voting, mixture of experts, stacking and cascading. Combined classifiers have an error percentage less than ind...

Consider the problem of visualizing high-dimensional data that has been categorized into various classes. Our goal in visualizing is to quickly absorb inter-class and intra-class relationships. Towards this end, class-preserving projections of the multidimensional data onto twodimensional planes, which can be displayed on a computer screen, are introduced. These class-preserving projections maintain the high-dimensional class structure, and are closely related to Fisher's linear discriminants. By displaying sequences of such two-dimensional projections and by moving continuously from one projection to the next, an illusion of smooth motion through a multidimensional display can be created. We call such sequences class tours. Furthermore, we overlay class-similarity graphs on our two-dimensional projections to capture the distance relationships in the original high-dimensional space. We illustrate

Learners based on different paradigms can be combined for improved accuracy. Each learning method assumes a certain model that comes with a set of assumptions which may lead to error if the assumptions do not hold. Learning is an ill-posed problem and with finite data each algorithm converges to a different solution and fails under different circumstances. Our previous experience with statistical and neural classifiers was that classifiers based on these paradigms do generalize differently, fail on different patterns and to a certain extent complement each other and thus we look for ways to combine them for higher accuracy. One way to get complementary classifiers is by using different input representations. The methods we investigate are voting, mixture of experts, stacking and cascading. We do experiments on three real-world applications: optical handwritten digit recognition, pen-based handwritten digit recognition and the estimation of road travel distances which is a regression pr...

Eager learners such as neural networks, decision trees, and nave Bayes classifiers construct a single model from the training data before observing any test set instances. In contrast, lazy learners such as Knearest neighbor consider a test set instance before they generalize beyond the training data. This allows making predictions from only a specific selection of instances most similar to the test set instance which has been shown to outperform eager learners on a number of problems. However, lazy learners require the storing and querying of the entire training data set for each instance which is unsuitable for the large amounts of data typically found in many applications. We introduce and illustrate the benefits of an example of semi-lazy learning that combines clustering and classification models and has the benefits of eager and lazy learners. We propose dividing the instances using cluster-based segmentation and then using an eager learner to build a classification model for each cluster. This has the effect of dividing the instance space into a number of distinct regions and building a local model for each. Our experiments on UCI data sets show clustering the data into segments then building classification models using a variety of eager learners for each segment often results in a greater overall cross-validated accuracy than building a single model or using a pure lazy approach such as K-Nearest-Neighbor. We can also consider our approach to semi-lazy learning as an example of the divide and conquer (DAC) strategy used in many scientific fields to divide a complex problem into a set of simpler problems. Finally, we find that the misclassified instances are more likely to be outliers with respect to the clustering segmentation.

In previous works, a time series classification system has been presented. It is based on boosting very simple classifiers, formed only by one literal. The used literals are based on temporal intervals.

In previous works, a time series classification system has been presented.

The main idea behind this project is that the pattern classification process can be enhanced by taking into account the geometry of class structure in datasets of interest. In contrast to previous work in the literature, this research not only develops a measure of discernibility of individual patterns but also consistently applies it to various stages of the classification process. The applications of the discernibility concept cover a wide range of issues from pre-processing to the actual classification and beyond that. Specifically, we apply it for: (a) finding feature subsets of similar classification quality (applicable in diverse ensembles), (b) feature selection, (c) data reduction, (d) reject option, and (e) enhancing the k-NN classifier. Also, a number of auxiliary algorithms and measures are developed to facilitate the proposed methodology. Experiments have been carried out using datasets of the University of California at Irvine (UCI) repository. The experiments provide numerical evidence that the proposed approach does improve the performance of various classifiers. This, together with its simplicity renders it a novel,