Feature Selection in Data-Mining for Genetics Using Genetic Algorithm

Abstract

We discovered genetic features and environmental factors which were involved in multifactorial diseases. To exploit the massive data obtained from the experiments conducted at the General Hospital, Chennai, data mining tools were required and we proposed a 2-Phase approach using a specific genetic algorithm. This heuristic approach had been chosen as the number of features to consider was large (upto 3654 for biological data under our study). Collected data indicated for pairs of affected individuals of a same family their similarity at given points (locus) of their chromosomes. This was represented in a matrix where each locus was represented by a column and each pairs of individuals considered by a row. The objective was first to isolate the most relevant associations of features and then to class individuals that had the considered disease according to these associations. For the first phase, the feature selection problem, we used a genetic algorithm (GA). To deal with this very specific problem, some advanced mechanisms had been introduced in the genetic algorithm such as sharing, random immigrant, dedicated genetic operators and a particular distance operator had been defined. Then, the second phase, a clustering based on the features selected during the previous phase, will use the clustering algorithm k-means.