Ryan G. Benton

Soft computing approach to steganalysis of LSB embedding in digital images

Steganography methods embed hidden data in digital images to provide a means of secret communication. Steganalysis methods detect such embedded data based on statistical analysis of a digital image. We experimented with using decision trees for detecting messages hidden in the least significant bit plane of an image and compared their results to those obtained using multilayered feedforward neural networks.

TRARM-RelSup: targeted rare association rule mining using itemset trees and the relative support measure

The goal of association mining is to find potentially interesting rules in large repositories of data. Unfortunately using a minimum support threshold, a standard practice to improve the association mining processing complexity, can allow some of these rules to remain hidden. This occurs because not all rules which have high confidence have a high support count. Various methods have been proposed to find these low support rules, but the resulting increase in complexity can be prohibitively expensive. In this paper, we propose a novel targeted association mining approach to rare rule mining using the itemset tree data structure (aka TRARM-RelSup). This algorithm combines the efficiency of targeted association mining querying with the capabilities of rare rule mining; this results in discovering a more focused, standard and rare rules for the user, while keeping the complexity manageable.

Data fusion and feature selection for Alzheimer's diagnosis

The exact cause of Alzheimers disease is unknown; thus, ascertaining what information is vital for the purpose of diagnosis, whether human or automated, is difficult. When conducting a diagnosis, one approach is to collect as much potentially relevant information as possible in the hopes of capturing the important information; this is the Alzheimers Disease Neuroimaging Initiative (ADNI) adopted approach. ADNI collects different clinical, image-based and genetic information related to Alzheimers disease. This study proposes a methodology for using ADNIs data. First, a series of support vector machines is constructed upon nine data sets. Five are the results of clinical tests and the other four are features derived from positron emission tomography (PET) imagery. Next, the SVMs are fused together to determine the final clinical dementia rating of a patient: normal or abnormal. In addition, the utility of applying feature selection methods to the generated PET feature data is demonstrated.

Exploitation of 3D stereotactic surface projection for predictive modelling of Alzheimer's disease

Alzheimers Disease (AD) is one major cause of dementia. Previous studies have indicated that the use of features derived from Positron Emission Tomography (PET) scans lead to more accurate and earlier diagnosis of AD, compared to the traditional approaches that use a combination of clinical assessments. In this study, we compare Naive Bayes (NB) with variations of Support Vector Machines (SVMs) for the automatic diagnosis of AD. 3D Stereotactic Surface Projection (3D-SSP) is utilised to extract features from PET scans. At the most detailed level, the dimensionality of the feature space is very high. Hence we evaluate the benefits of a correlation-based feature selection method to find a small number of highly relevant features; we also provide an analysis of selected features, which is generally supportive of the literature. However, we have also encountered patterns that may be new and relevant to prediction of the progression of AD.

Min-Max itemset trees for dense and categorical datasets

The itemset tree data structure is used in targeted association mining to find rules within a users sphere of interest. In this paper, we propose two enhancements to the original unordered itemset trees. The first enhancement consists of sorting all nodes in lexical order based upon the itemsets they contain. In the second enhancement, called the Min-Max Itemset Tree, each node was augmented with minimum and maximum values that represent the range of itemsets contained in the children below. For demonstration purposes, we provide a comprehensive evaluation of the effects of the enhancements on the itemset tree querying process by performing experiments on sparse, dense, and categorical datasets.

Procedural generation of stylized 2D maps

Outdoor worlds are often the setting for games and game worlds are often accompanied by a stylized version of the world drawn by an artist as a 2D map. Procedurally generating the terrain allows games to have a higher replay value. A limitation of procedural terrain generation is an artistic map of the terrain cannot be created by an artist beforehand. We propose an algorithm for generating a stylized 2D map from a simple procedurally generated 2D basis map. Our algorithm could be used in a game to generate stylized maps at execution time or in an offline application to serve as a guideline for an artist.

Detecting adverse drug effects using link classification on twitter data

Adverse drug events (ADEs) are among the leading causes of death in the United States. Although many ADEs are detected during pharmaceutical drug development and the FDA approval process, all of the possible reactions cannot be identified during this period. Currently, post-consumer drug surveillance relies on voluntary reporting systems, such as the FDAs Adverse Event Reporting System (AERS). With an increase in availability of medical resources and health related data online, interest in medical data mining has grown rapidly. This information coupled with online conversations of people which involve discussions about their health provide a substantial resource for the identification of ADEs. In this work, we propose a method to identify adverse drug effects from tweets by modeling it as a link classification problem in graphs. Drug and symptom mentions are extracted from the tweet history of each user and a drug-symptom graph is built, where nodes represent either drugs or symptoms and edges are labelled positive or negative, for desired or adverse drug effects respectively. A link classification model is then used to identify negative edges i.e. adverse drug effects. We test our model on 864 users using 10-fold cross validation with Siders dataset as ground truth. Our model was able to achieve an F-Score of 0.77 compared to the best baseline model with an F-Score of 0.58.

Framework for Real-Time Event Detection using Multiple Social Media Sources

Information about events happening in the real world are generated online on social media in real-time. There is substantial research done to detect these events using information posted on websites like Twitter, Tumblr, and Instagram. The information posted depends on the type of platform the website relies upon, such as short messages, pictures, and long form articles. In this paper, we extend an existing real-time event detection at onset approach to include multiple websites. We present three different approaches to merging information from two different social media sources. We also analyze the strengths and weaknesses of these approaches. We validate the detected events using newswire data that is collected during the same time period. Our results show that including multiple sources increases the number of detected events and also increase the quality of detected

Detection of event onset using Twitter.

Social Media generates information about news and events in real-time. Given the vast amount of data available and the rate of information propagation, reliably identifying events is a challenge. Most state-of-the-art techniques are post hoc techniques that detect an event after it happened. Our goal is to detect onset of an event as it is happening using the user-generated information from Twitter streams. To achieve this goal, we use a discriminative model to identify change in the pattern of conversations over time. We use a topic evolution model to find credible events and eliminate random noise that is prevalent in many of the event detection models. The simplicity of the proposed model allows detect events quickly and efficiently, permitting discovery of events within minutes from the start of conversation about those conversations on Twitter. Our model is evaluated on a large-scale Twitter corpus to detect events in real-time. The proposed model is tested on other datasets to detect change over longer periods of time. The results indicate we can detect real events, within 3 to 8 minutes of it first appearing, with a lower degree of noise compared to other methods.

A Comprehensive Granular Model for Decision Making with Complex

This chapter describes a comprehensive granular model for decision making with complex data. This granular model first uses information decomposition to form a horizontal set of granules for each of the data instances. Each granule is a partial view of the corresponding data instance; and aggregately all the partial views of that data instance provide a complete representation for the instance. Then, the decision making based on the original data can be divided and distributed to decision making on the collection of each partial view. The decisions made on all partial views will then be aggregated to form a final global decision. Moreover, on each partial view, a sequential M+1 way decision making (a simple extension of Yao’s 3-way decision making) can be carried out to reach a local decision. This chapter further categorizes stock price predication problem using the proposed decision model and incorporates the MLVS model for biological sequence classification into the proposed decision model. It is suggested that the proposed model provide a general framework to address the complexity and volume challenges in big data analytics.