Accurate classification of data sets is an important phenomenon for many applications. While multi-dimensionality to a certain point contributes to the classification performance, after a point, incorporating more attributes degrades the quality of the classification. In a pattern classification problem, by determining and excluding the least effective attribute(s) the performance of the classification is likely to improve. The task of the elimination of the least effective attributes in pattern classification is called "data dimensionality reduction (DDR)". DDR using Genetic Algorithms (DDR-GA) aims at discarding the less useful dimensions and re-organizing the data set by means of genetic operators. We show that a wise selection of the initial population improves the performance of the DDR-GA considerably and introduce a method to implement this approach. Our approach focuses on using information obtained a priori for the selection of initial chromosomes. Our work then compares the performance of the GA initiated by a randomly selected initial population to the performance of the ones initiated by a wisely selected one. Furthermore, the results indicate that our approach provides more accurate results compared to the purely random one in a reasonable amount of time.