Dependence Of Classification Accuracy Research Articles

Dynamical decision making in a genetic perceptron

Decision making is an essential element of cell functioning, which determines milestones of its evolution including differentiation, apoptosis and possible transition to cancerous state. Recently the concept of stochastic resonance in decision making (SRIDM) was introduced, demonstrated and explained using a synthetic genetic classifier circuit as an example. It manifests itself as a maximum in the dependence of classification accuracy upon noise intensity, and was caused by the concurrent action of two factors, both coarsening the classification accuracy by themselves, but found to extenuate the effect of each other: perturbation of classifier threshold and additive noise in classifier inputs. In the present work we extend the SRIDM concept to dynamical decision making, in which a classifier keeps track of the changeable input. We reproduce the stochastic resonance effect caused by noise and threshold perturbation, and demonstrate a new mechanism of SRIDM, which is associated with bistability and not connected with threshold perturbation.

New tools for evaluating LQAS survey designs

Lot Quality Assurance Sampling (LQAS) surveys have become increasingly popular in global health care applications. Incorporating Bayesian ideas into LQAS survey design, such as using reasonable prior beliefs about the distribution of an indicator, can improve the selection of design parameters and decision rules. In this paper, a joint frequentist and Bayesian framework is proposed for evaluating LQAS classification accuracy and informing survey design parameters. Simple software tools are provided for calculating the positive and negative predictive value of a design with respect to an underlying coverage distribution and the selected design parameters. These tools are illustrated using a data example from two consecutive LQAS surveys measuring Oral Rehydration Solution (ORS) preparation. Using the survey tools, the dependence of classification accuracy on benchmark selection and the width of the ‘grey region’ are clarified in the context of ORS preparation across seven supervision areas. Following the completion of an LQAS survey, estimation of the distribution of coverage across areas facilitates quantifying classification accuracy and can help guide intervention decisions.

Detection and classification of intrusions and faults using sequences of system calls

This paper investigates the use of sequences of system calls for classifying intrusions and faults induced by privileged processes in Unix. Classification is an essential capability for responding to an anomaly (attack or fault), since it gives the ability to associate appropriate responses to each anomaly type. Previous work using the well known dataset from the University of New Mexico (UNM) has demonstrated the usefulness of monitoring sequences of system calls for detecting anomalies induced by processes corresponding to several Unix Programs, such as sendmail, lpr, ftp, etc. Specifically, previous work has shown that the Anomaly Count of a running process, i.e., the number of sequences spawned by the process which are not found in the corresponding dictionary of normal activity for the Program, is a valuable feature for anomaly detection. To achieve Classification, in this paper we introduce the concept of Anomaly Dictionaries, which are the sets of anomalous sequences for each type of anomaly. It is verified that Anomaly Dictionaries for the UNM's sendmail Program have very little overlap, and can be effectively used for Anomaly Classification. The sequences in the Anomalous Dictionary enable a description of Self for the Anomalies, analogous to the definition of Self for Privileged Programs given by the Normal Dictionaries. The dependence of Classification Accuracy with sequence length is also discussed. As a side result, it is also shown that a hybrid scheme, combining the proposed classification strategy with the original Anomaly Counts can lead to a substantial improvement in the overall detection rates for the sendmail dataset. The methodology proposed is rather general, and can be applied to any situation where sequences of symbols provide an effective characterization of a phenomenon.

ARTMAP Neural Networks for Multispectral Image Classification

This paper presents an analysis of performance of several types of the ARTMAP neural network. The performance of the networks is analyzed in the task of classification of satellite images obtained by remote sensing. The analysis is concentrated on the dependence of classification accuracy on the difference in cluster type preferably identified by each of the classifiers. Three types of ARTMAP classifier are compared: fuzzy ARTMAP, Gaussian ARTMAP, and Extended Gaussian ARTMAP The main difference among these classifiers is in the way they determine/represent individual clusters in feature space. Best results are obtained for Extended Gaussian ARTMAP, a modification of the Gaussian ARTMAP neural network that preferably identifies Gaussian-distributed clusters.

Computational Modeling For The Study Of Multispectral Sensor Systems And Concepts

A computational model of the deterministic and stochastic processes involved in remote sensing is being developed as a tool for studying multispectral sensor systems and concepts. The goal is to improve the efficiency of sensor systems for routine worldwide monitoring of earth resources and the environment. Preliminary computational results are presented for simple models of the natural variability of atmospheric radiative transfer and surface reflectance. These results illustrate the dependence of classification accuracy on the selection of sensor spectral channels and data processing algorithms.