A Conv-LSTM Approach of Financial Pattern Prediction with Labeling from Feature-Based DTW with Pattern Rule

Equipped with financial market data labeled with DTW and Pattern Rule label showing the likelihood of corresponding sequences as specific financial pattern, a financial pattern prediction model can be developed by training the labeled data, to predict the probability of pattern formation in the future. In this paper, we applied a Seq2Seq deep learning model using Conv-LSTM to predict the probability of sequences forming the shape of 22 financial pattern in the future from one-time-step- to five-timestep-ahead. Such model is comprised of a Conv-LSTM regression model in predicting DTW and an imbalanced Conv-LSTM classification in predicting Pattern Rule. Massive technical details regarding Seq2Seq modeling were discussed, cleared and settled. Innovative reporting methodologies were developed to address model performance in multi-dimensional perspective. The final deliverable, Financial Pattern Prediction Model achieved high accuracy of predicting true future pattern formations, confirmed from not only labeled data, but also human perception