Overview / Usage

In short, this is a stock prediction model in Python. While most models I have developed were for personal use, and run locally, I attempted to deploy a model to production, using the web.py library, which has been employed by popular websites such as Reddit. Currently in the test phase, the demo is available at http://stocksify.ml

Methodology / Approach

With the model focused on using momentum as an indicator of future price. Our Python code uses scikit-learn and Keras to develop an LSTM model, as well as an RBF regression model. The third model is a simply long-term linear regression. The end advice served by the app utilizes and ensemble method based on the three models in deployment. Some drawbacks of this approach are that we must have an odd number of models, or else there may be ties, in which case we do not have a definitive result for the user.

Future considerations:

• Tracking public perception (applying sentiment analysis to news) to predict future prices.
• Using Q-Learning to develop a policy for trading based on price history.
• Using a sequential model such as HMM for long-term trading.
• Using CSP to make trading decisions under special constraints, like a trading budget.
• Adjusting degrees of certainty with a higher number of models, eg at least 75% concensus needed to buy a stock.

Technologies Used

I used the web.py and various SciPy libraries for back-end. This involved processing user requests, making predictions with my regression-based machine learning model, and serving those predictions to an HTML web page, rendered by the web.py library.

HTML, CSS were used to design the front end, as seen in the demo (see GitHub repo).

Repository

https://github.com/abhi1345/stocksify