A machine learning framework combining discrete wavelet transform preprocessing, continuous-trend labeling, and a temporal convolutional network achieved an average test accuracy of 72.85 percent in predicting stock price trends across a public dataset of 400 stocks. The architecture, designated DCTTCN, outperformed a comparative DAELM model by an absolute improvement of 0.96 percentage points, demonstrating competitive classification performance under an adopted continuous-trend definition.
According to the study, effective stock-trend prediction requires a clear definition of the target and an appropriate representation of price dynamics. To establish this representation, the DCTTCN framework applies continuous-trend labeling derived from turning points and a relative-price-change threshold. Under this labeling methodology, the defined trend intervals can include local price movements that occur in the opposite direction.
The system integrates wavelet-based representation construction with temporal modeling to classify market trends. Following initial discrete wavelet transform preprocessing, a temporal convolutional network learns temporal patterns from historical closing-price windows. The temporal convolutional network is then tasked with predicting the continuous-trend interval labels from these historical price sequences.
Performance was measured against the DAELM baseline across the 400-stock dataset, where DAELM recorded an average test accuracy of 71.89 percent compared to 72.85 percent for DCTTCN. The authors further conducted a trading-strategy evaluation to illustrate the practical application of the predicted labels in financial decision-making.
Subject of Research: Continuous stock-trend prediction using wavelet transforms and temporal convolutional networks
Article Title: Wavelet transform in conjunction with temporal convolutional networks for prediction of stock continuous trend
Article References: Wu, D. (2026). Wavelet transform in conjunction with temporal convolutional networks for prediction of stock continuous trend. International Journal of Data Science and Analytics, 22(1), Article 382. https://doi.org/10.1007/s41060-026-01358-5
Image Credits: AI Generated
DOI: 10.1007/s41060-026-01358-5
Keywords: DCTTCN framework, discrete wavelet transform, temporal convolutional network, stock-trend prediction, continuous-trend labeling, turning points, relative-price-change threshold, historical closing-price windows, price dynamics, DAELM baseline, 400-stock dataset, trading-strategy evaluation, stock-trend classification, wavelet-based representation, temporal pattern learning, trend interval labels
News Source: Scienmag.
