CS 631 – Research Trends in Artificial Intelligence Project Proposal Project Title Group Number Group Members Abstract Role of Neural Networks in Stock market prediction 10 Student Number 2005-02-0062 2005-02-0113 Name Hinna Aftab Mustafa Mahmood Stock price variation is an example of a multidimensional but partially-deterministic phenomenon to an extent. Neural networks are being used to predict stock prices based on a large number of factors including historic prices of other stocks and trends in various economic indicators. We would be analyzing the crucial role of neural networks in stock market prediction. Artificial Neural Networks (ANN) attempts to replicate the working of biological neural systems in a less complex manner. The nodes in the network behave like neurons in a biological system. All nodes in the network have certain inputs and output lines and every output line is connected to the input lines of another node. In this layered networks, the nodes are linked by a weighted connection to every node in the preceding layer. The nodes can be in one of the two states, active or inactive. The decision about the state of the node is a function of the weights of the input connections to adjacent nodes and the states of the adjacent nodes. Thus, the structure of the ANN may be organized into an input layer, an output layer and numerous intermediate layers. The input to the ANN is in a certain format that activates certain nodes, and the ANN produces an output pattern based on the weights of the connections. The learning takes place by comparing the resulting output pattern to the required output and then adjusting the weights of the connections accordingly. The intermediate and output layers determine their state based on the weights of the connections between nodes during their calculation iterations. The output pattern which is a set of active and inactive nodes is externally corrected for errors, while the error for intermediate nodes is determined by internal calculations during the learning iterations. If the error is positive, the weight is then redistributed to the active input lines strengthening the connections to the active nodes below. But if the error is negative, the weight is redistributed to the inactive input lines reducing the strength of the connections to the active nodes below. In this process, the weights of the output and intermediate layers are adjusted for error. The learning process comprises of the adjustment of the weights. As for stock market prediction, the ANN aims to assigns weights in a manner where the output patterns being generated are an approximately accurate measure of actual stock prices of the following days based on the stock price of previous day. Our project aims to cover the role of neural networks in stock market prediction in depth. We intend to study the ability of the ANN to utilize the knowledge stored in the connection weights to forecast stock prices, and its success to date.