Darryl N. Davis

A multi-agent decision support system for stock trading

A distributed problem solving system can be characterized as a group of individual cooperating agents running to solve common problems. As dynamic application domains continue to grow in scale and complexity, it becomes more difficult to control the purposeful behavior of agents, especially when unexpected events may occur. This article presents an information and knowledge exchange framework to support distributed problem solving. From the application viewpoint the article concentrates on the stock trading domain; however, many presented solutions can be extended to other dynamic domains. It addresses two important issues: how individual agents should be interconnected so that their resources are efficiently used and their goals accomplished effectively; and how information and knowledge transfer should take place among the agents to allow them to respond successfully to user requests and unexpected external situations. The article introduces an architecture, the MASST system architecture, which supports dynamic information and knowledge exchange among the cooperating agents. The architecture uses a dynamic blackboard as an interagent communication paradigm to facilitate factual data, business rule, and command exchange between cooperating MASST agents. The critical components of the MASST architecture have been implemented and tested in the stock trading domain, and have proven to be a viable solution for distributed problem solving based on cooperating agents.

Addressing the Class Imbalance Problem in Medical Datasets

A well balanced dataset is very important for creating a good prediction model. Medical datasets are often not balanced in their class labels. Most existing classification methods tend to perform poorly on minority class examples when the dataset is extremely imbalanced. This is because they aim to optimize the overall accuracy without considering the relative distribution of each class. In this paper we examine the performance of over-sampling and under-sampling techniques to balance cardiovascular data. Well known over-sampling technique SMOTE is used and some under-sampling techniques are also explored. An improved under sampling technique is proposed. Experimental results show that the proposed method displays significant better performance than the existing methods.

Reactive and Motivational Agents: Towards a Collective Minder

This paper explores the design and implementation of a societal arrangement of reflexive and motivational agents which will act as the building blocks for a more abstract agent within which the current agents act as distributed dynamic processing nodes. We contest that reactive, deliberative and other behaviours are required in complete (intelligent) agents. We provide some architectural considerations on how these differing forms of behaviours can be cleanly integrated and relate that to a discussion on the nature of motivational states and the mechanisms used for making decisions.

Reliability of Cephalometric Analysis Using Manual and Interactive Computer Methods

This study compares the results of cephalometric analyses using manual and interactive computer graphics methods. Results are statistically in favour of the interactive computer system. This study provides a basis for ongoing research into alternative methods of cephalometric analyses. such as digitization and automatic landmark identification using sophisticated computer vision systems.

Control States and Complete Agent Architectures

This paper presents a developing concept of mind defined in terms of external and internal niches. This perspective on mind is described primarily in terms of the niche space of control states and the design space of processes that may support such phenomena. A developing agent architecture, that can support motivation and other control states associated with mind, is presented. Different aspects of agent research are discussed in terms of three categories of agents. Each agent category is characterized primarily in terms of their task‐related competencies and internal behaviors and discussed in terms of our taxonomy of control states. The concept of complete agents is then introduced. Goals are described in terms of their generation across a number of computational layers. Experimental analysis is provided on how these differing forms of behaviors can be cleanly integrated. This leads into a discussion on the nature of motivational states and the mechanisms used for making decisions and managing the sometimes‐competitive nature of processes internal to a complex agent. The difficulty of evaluating complete agents is discussed from a number of perspectives. The paper concludes by considering future directions related to the computational modeling of emotions and the concept of synthetic mind.

Machine Learning Based Missing Value Imputation Method for Clinical Datasets

Missing value imputation is one of the biggest tasks of data pre-processing when performing data mining. Most medical datasets are usually incomplete. Simply removing the incomplete cases from the original datasets can bring more problems than solutions. A suitable method for missing value imputation can help to produce good quality datasets for better analysing clinical trials. In this paper we explore the use of a machine learning technique as a missing value imputation method for incomplete cardiovascular data. Mean/mode imputation, fuzzy unordered rule induction algorithm imputation, decision tree imputation and other machine learning algorithms are used as missing value imputation and the final datasets are classified using decision tree, fuzzy unordered rule induction, KNN and K-Mean clustering. The experiment shows that final classifier performance is improved when the fuzzy unordered rule induction algorithm is used to predict missing attribute values for K-Mean clustering and in most cases, the machine learning techniques were found to perform better than the standard mean imputation technique.

Computational architectures for intelligence and motivation

This paper presents an overview of five years work into computational architectures for intelligence and motivation. The starting point for this research is a generic computational architecture arising from the study of the control state perspective to mind. Variations on this architecture have been used to investigate questions about the nature of autonomy, emergence, belief management and motivation (and other control states) in synthetic agents. A number of different domains have been used, for example simple a-life environments, simulated robotic factories, five-a-side football, Tileworld and the game of Go. Due to the nature and scope of the work not all issues were addressed in any one domain. This raises questions about such endeavours. Given that it may be necessary to apportion fundamental research questions across different projects, can generalised conclusions be justifiably made from this division of research? The suggestion made here is that any such conclusions can only be justified in the light of subsequent integrative research. Current research directions are described in terms of these integrative questions.

Detection and characterization of carboniferous foraminifera for content-based retrieval from an image database

Carboniferous Foraminifers are a specific type of microfossil which are manifest in plane sections of rock and are used by geologists for dating rock samples. The images contain a high degree of visual noise and currently must be interpreted by human experts. We are studying the classification problem in the context of intelligent image databases. Here we present a technique for automatic identification of microfossil structures and for classification of the structures according to which type of 3-D section they represent. This is achieved by using: (1) A specialized filter to detect local curves in the gray level image data; and (2) Hough transform processing of the resulting feature point vectors. An interesting aspect of our approach is that the processing of the features is not embedded in a program but is instead specified using a visual query language. This allows us to experiment quickly with different types of grouping criteria. The detection performance of our system is comparable with that of a trained geologist. We store the information obtained in a database together with the raw image data. The system can then present the user with only those images which contain structures of interest.

Mining frequent biological sequences based on bitmap without candidate sequence generation

Biological sequences carry a lot of important genetic information of organisms. Furthermore, there is an inheritance law related to protein function and structure which is useful for applications such as disease prediction. Frequent sequence mining is a core technique for association rule discovery, but existing algorithms suffer from low efficiency or poor error rate because biological sequences differ from general sequences with more characteristics. In this paper, an algorithm for mining Frequent Biological Sequence based on Bitmap, FBSB, is proposed. FBSB uses bitmaps as the simple data structure and transforms each row into a quicksort list QS-list for sequence growth. For the continuity and accuracy requirement of biological sequence mining, tested sequences used during the mining process of FBSB are real ones instead of generated candidates, and all the frequent sequences can be mined without any errors. Comparing with other algorithms, the experimental results show that FBSB can achieve a better performance on both run time and scalability.

A comparative study of missing value imputation with multiclass classification for clinical heart failure data

Clinical data often contains missing values. Imputation is one of the best known schemes to overcome the drawbacks associated with missing values in data mining tasks. In this work, we compared several imputation methods and analyzed their performance when applied to different classification algorithms. A clinical heart failure data set was used in these experiments. The results showed that there is no universal imputation method that performs best for all classifiers. Some imputation-classification combinations are recommended for the processing of clinical heart failure data.