Financial forecasting using meta-cognitive fuzzy neural networks

• Main Author: Singh Akhand Pratap
• Other Authors: School of Computer Engineering
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; DRNTU::Engineering::Computer science and engineering::Computer systems organization::Special-purpose and application-based systems
• Online Access: http://hdl.handle.net/10356/59290
• Institution: Nanyang Technological University

Financial Markets have been attractive due to the thrill they provide through the profits or losses on the investments made in them. With the levels of achievements made in the field of Machine Learning and predictive mechanisms, there is a hope to finally break down the mystery of the financial markets through such machine learning techniques. The project tried to design a system to make investment decisions in the Stock Market using Type-2 Meta-Cognitive Fuzzy Neural Networks. Unlike simple learning mechanisms, meta-cognition attempts a selective learning approach. The primary meta-cognitive strategies are Sample deletion, Sample Learning and Sample Reserve. Following on the lines of Technical Analysis carried out by traders and investors on the historical prices of a company’s stock, the project makes use of various Financial Indicators to decide whether a particular instance of time is good for investment, disinvestment or to remain idle. The project in no manner tries to predict next day’s or the next few days’ stock prices, instead it predicts the output that is representative of what the future trend is going to be like. It does so, based on the expected lowest and highest ROI (per-day returns on the investment) made on that particular day. With a practical implementation, the results obtained aren’t phenomenally great, but on average, they do perform slightly better than the overall stock performance. The performance of the learning algorithm has been identified to be highly satisfactory, though the overall performance isn’t great. The setback identified in the project for the above stated problem is the way the prediction variables are calculated and the lack of automated mechanisms to adjust certain thresholds. Some potential ways to address the above issues could be usage of synthetic data through time-series models to train the model till the test date; usage of manual inferences drawn from the indicators to make judgements about the present day; usage of multiple machine learning models and combining the results from all; usage of dynamic feature selection mechanism to make improvements to the prediction; and usage of dynamic thresholds that are dependent on the market conditions. Other recommendations to target the problem of Financial Prediction in a more comprehensive manner could include addition of data from other sources like social media platforms, current affairs, global search queries, etc. to judge the public sentiment about a company. Another inclusion could have industry or market statistics or compilation of a single model of correlativeness of all the firms with each other that are operating within the market.