We leverage time-series data combined with pattern detection to detect buy and sell signals, combined with sentiment scores using from social media and news to confirm whether it is a positive move or negative move.
Gramian Angular Field (GAF) represents time series in a polar coordinate system instead of the typical Cartesian coordinates. Insides the GAF matrix( N x N), the element (i,j) is defined as either the cosine or the sine of the summation of the i th and j th angle.
GAF-CNN model accomplishes the buy/sell prediction purpose by implementing the pattern detection task. (see details in the Model Page)