This website requires JavaScript.

Stock Trend Prediction: A Semantic Segmentation Approach

Shima NabieeNader Bagherzadeh
Mar 2023
Market financial forecasting is a trending area in deep learning. Deeplearning models are capable of tackling the classic challenges in stock marketdata, such as its extremely complicated dynamics as well as long-term temporalcorrelation. To capture the temporal relationship among these time series,recurrent neural networks are employed. However, it is difficult for recurrentmodels to learn to keep track of long-term information. Convolutional NeuralNetworks have been utilized to better capture the dynamics and extract featuresfor both short- and long-term forecasting. However, semantic segmentation andits well-designed fully convolutional networks have never been studied fortime-series dense classification. We present a novel approach to predictlong-term daily stock price change trends with fully 2D-convolutionalencoder-decoders. We generate input frames with daily prices for a time-frameof T days. The aim is to predict future trends by pixel-wise classification ofthe current price frame. We propose a hierarchical CNN structure to encodemultiple price frames to multiscale latent representation in parallel usingAtrous Spatial Pyramid Pooling blocks and take that temporal coarse featurestacks into account in the decoding stages. Our hierarchical structure of CNNsmakes it capable of capturing both long and short-term temporal relationshipseffectively. The effect of increasing the input time horizon via incrementingparallel encoders has been studied with interesting and substantial changes inthe output segmentation masks. We achieve overall accuracy and AUC of %78.18and 0.88 for joint trend prediction over the next 20 days, surpassing othersemantic segmentation approaches. We compared our proposed model with severaldeep models specifically designed for technical analysis and found that fordifferent output horizons, our proposed models outperformed other models.