DYNAMIC SELF-ORGANISED NEURAL NETWORK INSPIRED BY THE IMMUNE ALGORITHM FOR FINANCIAL TIME SERIES PREDICTION AND MEDICAL DATA CLASSIFICATION

Artificial neural networks have been proposed as useful tools in time series analysis in a variety of applications. They are capable of providing good solutions for a variety of problems, including classification and prediction. However, for time series analysis, it must be taken into account that the variables of data are related to the time dimension and are highly correlated. The main aim of this research work is to investigate and develop efficient dynamic neural networks in order to deal with data analysis issues. This research work proposes a novel dynamic self-organised multilayer neural network based on the immune algorithm for financial time series prediction and biomedical signal classification, combining the properties of both recurrent and self-organised neural networks. The first case study that has been addressed in this thesis is prediction of financial time series. The financial time series signal is in the form of historical prices of different companies. The future prediction of price in financial time series enables businesses to make profits by predicting or simply guessing these prices based on some historical data. However, the financial time series signal exhibits a highly random behaviour, which is non-stationary and nonlinear in nature. Therefore, the prediction of this type of time series is very challenging. In this thesis, a number of experiments have been simulated to evaluate the ability of the designed recurrent neural network to forecast the future value of financial time series. The resulting forecast made by the proposed network shows substantial profits on financial historical signals when compared to the self-organised hidden layer inspired by immune algorithm and multilayer perceptron neural networks. These results suggest that the proposed dynamic neural networks has a better ability to capture the chaotic movement in financial signals. The second case that has been addressed in this thesis is for predicting preterm birth and diagnosing preterm labour. One of the most challenging tasks currently facing the healthcare community is the identification of preterm labour, which has important significances for both healthcare and the economy. Premature birth occurs when the baby is born before completion of the 37-week gestation period. Incomplete understanding of the physiology of the uterus and parturition means that premature labour prediction is a difficult task. The early prediction of preterm births could help to improve prevention, through appropriate medical and lifestyle interventions. One promising method is the use of Electrohysterography. This method records the uterine electrical activity during pregnancy. In this thesis, the proposed dynamic neural network has been used for classifying between term and preterm labour using uterine signals. The results indicated that the proposed network generated improved classification accuracy in comparison to the benchmarked neural network architectures