Classification and Preprocessing in the Stock Data

Abstract

In this paper we deal with the problem of assigning classes to the given market situation. We consider approach in which every market situation can be connected with one of the following decision classes: BUY, SELL or WAIT. Each of two classes: BUY and SELL can be assigned only on the basis of significant rises or drops of the given instrument. In all remaining cases WAIT class is assigned. Such approach allows to be independent of indicator values which nowadays are considered to have the significant prediction power. To achieve the goal we selected various stock instruments and with the use of the preprocessing and data discretization we generated decision tables for every considered datasets.

Furthermore, decision trees is built on the basis of generated decision tables. Decision trees are used in the process of classification of newly generated stock data. Presented approach is tested with the use of two independent sets: training set – used to built classifiers – decision classes, and test set – used to estimate accuracy of the generated decision trees. Finally we refer results to other approach in which forex data were used.