Stock Trend Analysis and Prediction Algorithm

USC EE Master Term Paper: Developed a stock trend analysis algorithm based on time-series clustering and trend classification to predict stock price movement.

* The algorithm utilizes windowing and a gradient-based pre-clustering step to provide initial centroids for K-means clustering.
* Classification and prediction of testing data are based on the shortest Euclidean distance to the resulting trend-labeled centroids (K groups).
* Achieved effective prediction results, particularly for the overall trend movement of Apple (AAPL) and Ebay (EBAY) stocks, with parameters set at a training window of 15 days, a testing window of 10 days, and K=3 clusters.