A model fusion method based on multi-source heterogeneous data for stock trading signal prediction

In the prediction of turning points (TPs) time series, improved model integrating piecewise linear representation and weighted support vector machine (IPLR-WSVM) has achieved good performance. However, due to single data source limitation algorithm, IPLR-WSVM encountered challenges in profitability. this paper, a fusion method based on multi-source heterogeneous different learning algorithms is proposed for TPs (MF-MSHD). Multi-source include unstructured structured information with granularities. RF, WSVM, BPNN, GBDT, LSTM are selected be algorithms. The differences among meta-models constructed by inputs as much possible, rule designed determine final TPs. Moreover, generated characteristics individual stock. For sentiment analysis, more accurate dictionary stock market comments established. Specifically, fine-grained introduced jointly trading moment. level proposal improves accuracy profitability, also outperforms composite indexes. Experimental results show that profit rate randomly stocks MF-MSHD reaches 0.5172, while highest value 0.2841 meta-model 0.0992 buy hold strategy, respectively. other indicators including modified. Compared increases 0.1648, 0.4051, 0.3397 Shanghai Composite Index, Shenzhen CSI 300 shows higher profitability signal prediction.