|dc.description.abstract||Stock Price Prediction is a challenging, interesting and potentially very profitable task to
carry out. The task has great dependence on economic phenomena, history, politics, the
media, hype and even some psychology. Experts in the relevant fields have developed
elaborate formal techniques and formulas that one may use in trying to carry out the task.
On many a case these experts would not agree on anyone of the ways as being the best to
go about predicting stock market figures. A lot of investors utilize intuition with little
reference to the hard facts and figures. Utilizing a data mining methodology, CRlSP
DM, and the artificial intelligence technique of Neural Networks, this project sets out to
provide yet another way of making stock price predictions specifically at the Nairobi
Stock Exchange. The neural networks are trained using publicly available historical data,
with the intention of deploying the networks to learn any explicit and the not so explicit
relationships that may exist in these data. Extensive tests are carried out to compare the
performance of different network topologies and parameter settings and leading to the
determination of a good network architecture for this application.
Tests carried out have brought about the construction of models with high prediction
. accuracy, including one with a mean deviation error rate of ±4.08%.||en