Adrian Gepp

A Comparative Analysis of Decision Trees Vis-à-vis Other Computational Data Mining Techniques in Automotive Insurance Fraud Detection

The development and application of computational data mining techniques in nancial fraud detection and business failure prediction has become a popular cross-disciplinary research area in recent times involv- ing nancial economists, forensic accountants and computational modellers. Some of the computational techniques popularly used in the context of - nancial fraud detection and business failure prediction can also be eectively applied in the detection of fraudulent insurance claims and therefore, can be of immense practical value to the insurance industry. We provide a compara- tive analysis of prediction performance of a battery of data mining techniques using real-life automotive insurance fraud data. While the data we have used in our paper is US-based, the computational techniques we have tested can be adapted and generally applied to detect similar insurance frauds in other countries as well where an organized automotive insurance industry exists.

A review of procedures to evolve quantum algorithms

There exist quantum algorithms that are more efficient than their classical counterparts; such algorithms were invented by Shor in 1994 and then Grover in 1996. A lack of invention since Grover’s algorithm has been commonly attributed to the non-intuitive nature of quantum algorithms to the classically trained person. Thus, the idea of using computers to automatically generate quantum algorithms based on an evolutionary model emerged. A limitation of this approach is that quantum computers do not yet exist and quantum simulation on a classical machine has an exponential order overhead. Nevertheless, early research into evolving quantum algorithms has shown promise. This paper provides an introduction into quantum and evolutionary algorithms for the computer scientist not familiar with these fields. The exciting field of using evolutionary algorithms to evolve quantum algorithms is then reviewed.

Predicting FTSE 100 Returns and Volatility Using Sentiment Analysis

We investigate the statistical and economic effect of positive and negative sentiment on daily excess returns and volatility in the FTSE 100 index, using business news articles published by the Guardian Media Group between 01/01/2000 and 01/06/2016. The analysis indicates that while business news sentiment derived from articles aimed at retail traders does not influence excess returns in the FTSE 100 index, it does affect volatility, with negative sentiment increasing volatility and positive sentiment reducing it. Further, an ETF‐based trading strategy based on these findings is found to outperform the naive buy‐and‐hold approach.

Automatically Generating and Solving Eternity II Style Puzzles

The Eternity II puzzle is an NP-complete problem. Prior researchers have generated data sets that are similar to the Eternity II problem. These data sets can be created in linear time, but this comes at the cost of easing the problem by introducing exploitable statistical features. The first contribution of this paper is a new method to generate data sets that are truly of Eternity II style. The second contribution is an Eternity II specific implementation of a constraint-satisfaction-problem style algorithm. Unlike most other published algorithms, this one has no form of look-ahead, filtering, forward checking, back jumping or k-consistency checks. Instead, it uses knowledge about the structure of the puzzle and the uniform distribution of edge colours. This approach is up to three orders of magnitude faster than previously published attempts.

The Effect of Sentiment on Stock Price Prediction

Accurately predicting stock prices is of great interest to both academics and practitioners. However, despite considerable efforts over the last few decades, it still remains an elusive challenge. For each of Australia’s 20 largest stocks, we build two neural network autoregressive (NNAR) models: one a basic NNAR model, and the other an NNAR model extended with sentiment inputs. By comparing the prediction accuracy of the two models, we find evidence that the inclusion of sentiment variables based on news articles and twitter sentiment can enhance the accuracy of the stock price prediction process.

Financial-distress prediction of Islamic banks using tree-based stochastic techniques

Financial distress is a socially and economically important problem that affects companies the world over. Having the power to better understand – and hence aid businesses from failing, has the potential to save not only the company, but also potentially prevent economies from sustained downturn. Although Islamic banks constitute a fraction of total banking assets, their importance have been substantially increasing, as their asset growth rate has surpassed that of conventional banks in recent years. The paper aims to discuss these issues.,This paper uses a data set comprising 101 international publicly listed Islamic banks to work on advancing financial distress prediction (FDP) by utilising cutting-edge stochastic models, namely decision trees, stochastic gradient boosting and random forests. The most important variables pertaining to forecasting corporate failure are determined from an initial set of 18 variables.,The results indicate that the “Working Capital/Total Assets” ratio is the most crucial variable relating to forecasting financial distress using both the traditional “Altman Z-Score” and the “Altman Z-Score for Service Firms” methods. However, using the “Standardised Profits” method, the “Return on Revenue” ratio was found to be the most important variable. This provides empirical evidence to support the recommendations made by Basel Accords for assessing a bank’s capital risks, specifically in relation to the application to Islamic banking.,These findings provide a valuable addition to the limited literature surrounding Islamic banking in general, and FDP pertaining to Islamic banking in particular, by showcasing the most pertinent variables in forecasting financial distress so that appropriate proactive actions can be taken.

Text Mining and Automation for Processing of Patient Referrals

BACKGROUND Various tasks within health care processes are repetitive and time-consuming, requiring personnel who could be better utilized elsewhere. The task of assigning clinical urgency categories to internal patient referrals is one such case of a time-consuming process, which may be amenable to automation through the application of text mining and natural language processing (NLP) techniques. OBJECTIVE This article aims to trial and evaluate a pilot study for the first component of the task-determining reasons for referrals. METHODS Text is extracted from scanned patient referrals before being processed to remove nonsensical symbols and identify key information. The processed data are compared against a list of conditions that represent possible reasons for referral. Similarity scores are used as a measure of overlap in terms used in the processed data and the condition list. RESULTS This pilot study was successful, and results indicate that it would be valuable for future research to develop a more sophisticated classification model for determining reasons for referrals. Issues encountered in the pilot study and methods of addressing them were outlined and should be of use to researchers working on similar problems. CONCLUSION This pilot study successfully demonstrated that there is potential for automating the assignment of reasons for referrals and provides a foundation for further work to build on. This study also outlined a potential application of text mining and NLP to automating a manual task in hospitals to save time of human resources.