This project sought to use rudimentary machine learning techniques to provide sound investment advice to the retail investor. Data on all actively-traded equities, sampled weekly from 2014 to 2018, was pulled from a Washington University Bloomberg Terminal. A random forest algorithm was run on this data to provide insight such as feature importance and ‘buy’ recommendations on individual equities. A k-means clustering algorithm was then be run to determine whether there is a particular subset of equities which was predicted to perform will. The results of this analysis are inconclusive. The random forest yielded unrealistically high prediction accuracy for test data and the K-means clustering yielded no discernible insight.
Download the full report here.