An Improved Multi-Objective Genetic Model for Stock Selection with Domain Knowledge - Technologies and Applications of Artificial Intelligence - page 68

In this study, we propose an improved MOGA method with some financial know-
ledge in criteria evaluation to make it more suitable for stock selection than our pre-
vious work [2]. We also employ stock datasets of twenty years of time in order to
conduct a more convincing statistical validation of the effectiveness of our approach.

3 Our Proposed Algorithm

This section describes our proposed algorithm for the construction of our stock scor-
ing model, optimized by the GA and the corresponding extension of our previous
model [2] in the context of multi-objective optimization. The algorithm is presented
as follows:

1. Devise a chromosome that consists of three portions ― the feature selection , the
stock sorting indicators and the feature weights . A chromosome is
represented by a binary coding scheme.

Fig. 1. Chromosome encoding

In Figure 1, is the number of fundamental variables (features). through
are defined for candidate features 1 through , where 1 or 0 corresponds to the fea-
ture being selected or not. through are defined as the sorting indicators, where
0 represents the higher the value of the variable, the better; and 1 represents the oppo-
site case. through are defined as the encoding of the set of weights . Fig-
ure 2 shows the detailed binary encoding for the weight of each individual variable,
where the value of (the weight for variable ) in Fig. 1 is encoded by loci
through .

Fig. 2. Detailed encoding of the weighting terms

The portion in the chromosome representing the genotype of weight is trans-
formed into the phenotype as follows:

y

,

(1)

where is the corresponding phenotype for the particular weight; and
are the minimum and maximum of the parameters; is the corresponding decimal
value, and is the length of the block used to encode the weight in the chromosome.