Non-parametric Classification Algorithm Research Articles

Mapping megacity growth with multi-sensor data

Abstract In our increasingly urbanized world, monitoring and mapping of urban growth and thereby induced land-use and land-cover change (LULCC) is of emergent importance. Remote sensing can reveal spatio-temporal growth trajectories of cities, which again allow a thorough understanding of the impacts of urbanization on ecosystems and ecosystem services. However, the mapping of urban areas remains one of the most challenging tasks of remote sensing data analysis. This paper presents an approach to map urban growth from multi-sensoral data, exemplified for the Dhaka megacity region in Bangladesh between 1990 and 2006. The approach is globally applicable and can facilitate regional urban growth maps in arbitrary complex and dynamic environments. Dhaka's densified urban landscape, its deltaic locality and the highly dynamic monsoon-related phenology call for a sophisticated analysis approach that is able to separate intra-annual land-cover variations from actual urbanization. Imagery from the Landsat series of satellites is a great asset for such an analysis due to its synoptic coverage of large urban areas as well as its unique historical archives. In our approach, we solve problems of spectral ambiguities and seasonal phenological dynamics through incorporating multi-temporal imagery for each monitoring year (1990, 2000, and 2006) and by extending the spectral feature space with synthetic aperture radar (SAR) data. The resulting datasets are heterogeneous and comprise measurements of unequal scaling. Non-parametric classification algorithms are required to delineate multi-modal and non-Gaussian class distributions of heterogeneous as well as temporally and spectrally complex land-cover classes of interest in such an extended feature space. We therefore used a support vector machine (SVM) classifier and post classification comparison to reveal spatio-temporal patterns of urban land-use and land-cover changes. An SVM based forward feature selection procedure allowed deriving in-depth information about the individual contribution of different input bands. Our methodology delineated relevant land-cover classes and resulted in overall accuracies better than 83% for all years considered. Change analysis unveiled a profound expansion of urban areas at the expense of prime agricultural areas and wetlands. During the 1990s, change was primarily characterized by a densification of urban fabric, whereas more recent changes included vast in-filling of low lying land and an extensive industrial sprawl into Dhaka's peri-urban areas. Our multi-sensoral and multi-temporal mapping approach allowed for delineating temporally dynamic LULCC, which again allowed for an insightful characterization of land system changes in the megacity region of Dhaka.

Mortgage Default: Classification Trees Analysis

We apply the powerful, flexible, and computationally efficient nonparametric Classification and Regression Trees (CART) algorithm to analyze real estate mortgage data. CART is particularly appropriate for our data set because of its strengths in dealing with large data sets, high dimensionality, mixed data types, missing data, different relationships between variables in different parts of the measurement space, and outliers. Moreover, CART is intuitive and easy to interpret and implement. We discuss the pros and cons of CART in relation to traditional methods such as linear logistic regression, nonparametric additive logistic regression, discriminant analysis, partial least squares classification, and neural networks, with particular emphasis on real estate. We use CART to produce the first academic study of Israeli mortgage default data. We find that borrowers’ features, rather than mortgage contract features, are the strongest predictors of default if accepting icbadli borrowers is more costly than rejecting “good” ones. If the costs are equal, mortgage features are used as well. The higher (lower) the ratio of misclassification costs of bad risks versus good ones, the lower (higher) are the resulting misclassification rates of bad risks and the higher (lower) are the misclassification rates of good ones. This is consistent with real-world rejection of good risks in an attempt to avoid bad ones.

Mortgage Default: Classification Trees Analysis

We apply the powerful, flexible, and computationally efficient nonparametric Classification and Regression Trees (CART) algorithm to analyze real estate mortgage data. CART is particularly appropriate for our data set because of its strengths in dealing with large data sets, high dimensionality, mixed data types, missing data, different relationships between variables in different parts of the measurement space, and outliers. Moreover, CART is intuitive and easy to interpret and implement. We discuss the pros and cons of CART in relation to traditional methods such as linear logistic regression, nonparametric additive logistic regression, discriminant analysis, partial least squares classification, and neural networks, with particular emphasis on real estate. We use CART to produce the first academic study of Israeli mortgage default data. We find that borrowers' features, rather than mortgage contract features, are the strongest predictors of default if accepting bad borrowers is more costly than rejecting good ones. If the costs are equal, mortgage features are used as well. The higher (lower) the ratio of misclassification costs of bad risks versus good ones, the lower (higher) are the resulting misclassification rates of bad risks and the higher (lower) are the misclassification rates of good ones. This is consistent with real-world rejection of good risks in an attempt to avoid bad ones.

Investigation of fiber-optic probe designs for optical spectroscopic diagnosis of epithelial pre-cancers.

The first objective of this study was to evaluate the performance of fluorescence spectroscopy for diagnosing pre-cancers in stratified squamous epithelial tissues in vivo using two different probe geometries with (1) overlapping versus (2) non-overlapping illumination and collection areas on the tissue surface. Probe (1) and probe (2) are preferentially sensitive to the fluorescence originating from the tissue surface and sub-surface tissue depths, respectively. The second objective was to design a novel, angled illumination fiber-optic probe to maximally exploit the depth-dependent fluorescence properties of epithelial tissues. In the first study, spectra were measured from epithelial pre-cancers and normal tissues in the hamster cheek pouch and analyzed with a non-parametric classification algorithm. In the second study, Monte Carlo modeling was used to simulate fluorescence measurements from an epithelial tissue model with the angled illumination probe. An unbiased classification algorithm based on spectra measured with probes (1) and (2), classified pre-cancerous and normal tissues with 78 and 94% accuracy, respectively. The angled illumination probe design provides the capability to detect fluorescence from a wide range of tissue depths in an epithelial tissue model. The first study demonstrates that fluorescence originating from sub-surface tissue depths (probe (2)) is more diagnostic than fluorescence originating from the tissue surface (probe (1)) in the hamster cheek pouch model. However in general, it is difficult to know a priori the optimal probe geometry for pre-cancer detection in a particular epithelial tissue model. The angled illumination probe provides the capability to measure tissue fluorescence selectively from different depths within epithelial tissues, thus obviating the need to select a single optimal probe design for the fluorescence-based diagnosis of epithelial pre-cancers.

The application of non-parametric techniques to solve classification problems in complex data sets in veterinary epidemiology – An example

Statistical classification problems are very common in veterinary epidemiology. Traditionally, parametric techniques such as logistic regression or discriminant analysis are used to analyse data sets that contain several classes of observations. However, characteristics of the data set such as high dimensionality, multicollinearity and non-homogeneity can make a data set unsuitable for parametric techniques. In this article, classification tree algorithms (ID3, C4.5, CHAID, CART) and artificial neural networks are suggested as non-parametric alternatives. Their application is illustrated using a field data set containing pig farms with 3 levels of respiratory disease prevalence. The performance of non-parametric classification algorithms is compared with results from multinomial logistic regression. None of the algorithms was significantly better than the others. The proportions of correctly classified farms were between 84% and 96%. However, the data set was small (86 observations), which created technical problems when using the artificial neural networks and multinomial logistic regression. The choice of statistical technique should therefore be based on the objectives of the study and the data set under consideration. Classification trees are well-suited for exploratory data analysis. They are easy to apply and worth considering.

Automatic visual inspection of surface mount solder joint defects

The ever increasing use of surface mount technology in printed circuit board assembly has considerably increased the importance of automated solder joint inspection. This paper documents our experience with the development of an experimental automated solder joint inspection system. Dominant solder joint defect classes are identified. Three-dimensional geometric models of these defects are created and used to test a non-parametric classification algorithm and optimal feature selection scheme. The classification algorithm and optimal feature selection scheme are then implemented on an experimental system using a laser line and vision system for solder joint data acquisition. The capability and limitations of this system are assessed through experimentation with actual solder joint defects.

Probabilistic neural network architecture for high-speed classification of remotely sensed imagery

In this article we discuss a neural network architecture (the probabilistic neural net or the PNN) that, to the best of our knowledge, has not previously been applied to remotely sensed data. The PNN is a supervised nonparametric classification algorithm as opposed to the Gaussian maximum likelihood classifier (GMLC). The PNN works by fitting a Gaussian kernel to each training point. The width of the Gaussian is controlled by a tuning parameter called the window width. If very small widths are used, the method is equivalent to the nearest neighbor method. For large windows, the PNN behaves like the GMLC. The basic implementation of the PNN requires no training time at all. In this respect it is far better than the commonly used backpropagation neural network (BPNN), which can be shown to the O( N 6) time for training where N is the dimensionality of the input vector. In addition, the PNN can be implemented in a feed-forward mode in hardware. The disadvantage of the PNN is that it requires all the training data to be stored. Some solutions to this problem are discussed in the article. Finally, we discuss the accuracy of the PNN with respect to the GMLC and the BPNN. The PNN is shown to be better than GMLC and not as good as the BPNN with regard to classification accuracy.