|
Abstract:
|
While most participants in a data science competition will not see their efforts rewarded by the obvious motivations, the difficult nature of the underlying problem causes many competitors to find a stronger motivation in the opportunity to develop a novel solution. Since obvious approaches to these problems do not work, successful results are based on the creativity of the competitors, their unique areas of expertise, and their ability to capture the less obvious features in the data. However, the goal of developing such a unique model, combined with the limited amount of time in which to work, greatly disrupts the normal routine of statistical modeling. Having participated in many data mining competitions, I have had the opportunity to notice some common techniques which allow competitors to build successful models, as well as understandable mistakes that lead to less successful results. By avoiding the traps that limit exploration and successfully shifting focus between the larger and smaller details, competitors can recognize important frameworks upon which to best drape and model the data.
|
