Perception of patterns and trends in financial markets using polar graph and contour mapping

In recent years, small and medium businesses across different industries have confronted the challenges of big data analysis since a huge amount of data being generated daily by their business activities. Most of these companies are unable to achieve efficient data analysis and decision-making based on such a high dimensional and voluminous data. Normally, the performance of analytic models will be limited if the business companies directly use the original data to train, verify and test their models. Therefore, to eliminate the complexity and computation of data analysis, the raw data requires an effective transformation to reduce the dimensionality of data. In this study, non-temporal data are transformed to a two-dimensional polar graph for further analysis. On the other hand, the temporal data combined with cross-sectional data are mapped to another two-dimensional contour graph that derived and sliced from their corresponding three-dimensional data profile. Both the transforming strategies are converted and fulfilled with various geometric shape descriptors and invariant moments, and three conventional machine-learning approaches are implemented to evaluate their predictive performance. In addition, an autoencoder neural network based on unsupervised learning algorithm is also employed to evaluate predictive accuracy in comparison with the conventional approaches. The experiment results suggested that the autoencoder neural network achieved the highest accuracy, and the rest approaches were considered worse than the below-chance accuracy.