Association rule learning is a machine learning approach aiming to find substantial relations among attributes within one or more datasets. We address the main problem of this technology, which is the excessive computation time and the memory requirements needed for the processing of discovering the association rules. Most of the literature pertaining to the association rules deals extensively with these issues as major obstacles, especially for very large databases. In this paper, we introduce a method that requires substantially lowers the run time and memory requirements in comparison to the methods presently in use (reduction from ${O(2}^m)$ to $O\left(2^{\frac{m}{2}}\right)$ in the worst case).

The study has two objectives. The first, to develop an earnings movement prediction model to help investors in their decision process, the second, to explore the potential of Recurrent Neural Networks (RNN) in financial statement analysis and present a detailed model for its application. RNN's two major advantages are: they do not make assumptions regarding the data and allow users to search whatever functional form best describes the underlying relationship between financial data and changes in earnings; they dynamically account for time-series behavior, earnings of a certain time period are not independent of earnings in previous time periods. The paper utilizes the newly mandated XBRL data, whose benefits are that it is freely available, easily accessible and is more timely than traditional databases. The use of RNN is validated in the results by providing a higher accuracy prediction than neural networks and logistic regression.