Richard J. Roiger

Gamma-Ray Burst Class Properties

Guided by the supervised pattern recognition algorithm C4.5 developed by Quinlan in 1986, we examine the three gamma-ray burst classes identified by Mukherjee et al. in 1998. C4.5 provides strong statistical support for this classification. However, with C4.5 and our knowledge of the BATSE instrument, we demonstrate that class 3 (intermediate fluence, intermediate duration, soft) does not have to be a distinct source population: statistical/systematic errors in measuring burst attributes combined with the well-known hardness/intensity correlation can cause low peak flux class 1 (high fluence, long, intermediate hardness) bursts to take on class 3 characteristics naturally. Based on our hypothesis that the third class is not a distinct one, we provide rules so that future events can be placed in either class 1 or class 2 (low fluence, short, hard). We find that the two classes are relatively distinct on the basis of Bands work in 1993 on spectral parameters α, β, and Epeak alone. Although this does not indicate a better basis for classification, it does suggest that different physical conditions exist for class 1 and class 2 bursts. In the process of studying burst class characteristics, we identify a new bias affecting burst fluence and duration measurements. Using a simple model of how burst duration can be underestimated, we show how this fluence duration bias can affect BATSE measurements and demonstrate the type of effect it can have on the BATSE fluence versus peak flux diagram.

How Sample Completeness Affects Gamma-Ray Burst Classification

Unsupervised pattern-recognition algorithms support the existence of three gamma-ray burst classes: class 1 (long, large-fluence bursts of intermediate spectral hardness), class 2 (short, small-fluence, hard bursts), and class 3 (soft bursts of intermediate durations and fluences). The algorithms surprisingly assign larger membership to class 3 than to either of the other two classes. A known systematic bias has been previously used to explain the existence of class 3 in terms of class 1; this bias allows the fluences and durations of some bursts to be underestimated, as recently shown by Hakkila et al. We show that this bias primarily affects only the longest bursts and cannot explain the bulk of the class 3 properties. We resolve the question of class 3s existence by demonstrating how samples obtained using standard trigger mechanisms fail to preserve the duration characteristics of small-peak flux bursts. Sample incompleteness is thus primarily responsible for the existence of class 3. In order to avoid this incompleteness, we show how a new, dual-timescale peak flux can be defined in terms of peak flux and fluence. The dual-timescale peak flux preserves the duration distribution of faint bursts and correlates better with spectral hardness (and presumably redshift) than either peak flux or fluence. The techniques presented here are generic and have applicability to the studies of other transient events. The results also indicate that pattern recognition algorithms are sensitive to sample completeness; this can influence the study of large astronomical databases, such as those found in a virtual observatory.

A Gamma-Ray Burst Database of BATSE Spectral Lag and Internal Luminosity Function Values

We present a database of spectral lags and internal luminosity function (ILF) measurements for gamma-ray bursts (GRBs) in the BATSE catalog. Measurements were made using 64 ms count rate data and are defined for various combinations of the four broadband BATSE energy channels. We discuss the processes used for measuring lags and ILF characteristics. We discuss the statistical and systematic uncertainties in measuring these attributes, as well as the role of temporal resolution in measuring lags and/or ILFs—these are particularly noticeable for GRBs belonging to the Short class. Correlative and clustering properties of the lag and ILF are examined, including the ability of these attributes to predict GRB time history morphologies. We conclude that the ILF and lag have great potential for studying GRB physics when used with other burst attributes.

A majority rules approach to data mining

Knowledge discovery in databases (KDD) offers a methodology for developing tools to extract meaningful knowledge from large volumes of data. We propose a generalized KDD model for supervised training. A main step in this process, data mining, involves the creation of a classification structure that is representative of the concept classes identified in the data set. Data mining incorporates learning which may be supervised or unsupervised and often uses statistical as well as heuristic (machine learning) techniques. Previous research has shown that different supervised models perform better under certain conditions. We tested the extent of overlap of instance classifications between five supervised models in two real world domains. Experimental results showed that in one domain all five models classified 75.8% of the instances identically, correct or incorrect. In the second domain, the corresponding figure was 63.3%. The amount of agreement between models can be used to help determine the nature of the domain and the applicability of a supervised learning approach. We extend the above experimental result and propose a multi model majority rules (MR) data mining technique to learn about the nature of a given domain. We conclude with directions for future work.

A tool for public analysis of scientific data

The scientific method encourages sharing data with other researchers to independently verify conclusions. Currently, technical barriers impede such public scrutiny. A strategy for offering scientific data for public analysis is described. With this strategy, effectively no requirements of software installation (other than a web browser) or data manipulation are imposed on other researchers to prepare for perusing the scientific data. A prototype showcasing this strategy is described.

Teaching an introductory course in data mining

The goal is to supply the participant with the tools to teach a course or unit about data mining and knowledge discovery. A basic understanding of the benefits and limitations of data mining as a problem-solving strategy will be offered. Several data mining techniques will be discussed. Prior knowledge about data mining and the knowledge discovery process is not necessary.

Unsupervised induction and gamma-ray burst classification

We use ESX, a product of Information Acumen Corporation, to perform unsupervised learning on a data set containing 797 gamma-ray bursts taken from the BATSE 3B catalog [5]. Assuming all attributes to be distributed log-normally, Mukherjee et al. [6] analyzed these same data using a statistical cluster analysis. Utilizing the logarithmic values for T90 duration, total fluence, and hardness ratio HR321 their results showed the instances formed three classes. Class I contained long/bright/intermediate bursts, class II consisted of short/faint/hard bursts and class III was represented by intermediate/intermediate/soft bursts. When ESX was presented with these data and restricted to forming a small number of classes, the two classes found by previous standard techniques [1] were determined. However, when ESX was allowed to form more than two classes, four classes were created. One of the four classes contained a majority of short bursts, a second class consisted of mostly intermediate bursts, and the final two...

A GRB tool shed

We describe the design of a suite of software tools to allow users to query Gamma Ray Burst (GRB) data and perform data mining expeditions. We call this suite of tools a shed (SHell for Expeditions using Datamining). Our schedule is to have a completed prototype (funded via the NASA AISRP) by February, 2002. Meanwhile, interested users will find a partially functioning tool shed at http:/grb.mankato.msus.edu.

Substance user MMPI-2 profiles : Predicting failure in completing treatment

Minnesota Multiphasic Personality Inventory-2 protocols were examined in an attempt to develop a model able to identify chemically-dependent patients likely not to complete treatment. MMPI-2 profiles of 173 patients (142 male) were analyzed using profile code types and a multiple analysis of variance. A chi-square showed that patients classified as neurotic were more likely to fail treatment. A MANOVA indicated that elevated T-scores on Scales 7 and 8 (p < .05) were related to noncompletion. Comparing these results with similar studies indicates that attempting to construct a predictive model based on a single objective measure may not be sufficient to determine outcome.

An Update on the GRB ToolSHED Project Status

The GRB ToolShed is an online suite of induction‐based machine learning and statistical tools designed for gamma‐ray burst classification and cluster analysis. The ToolSHED also includes a large preprocessed gamma‐ray burst database. We report on the current status of the ToolSHED.