Linear regression is a well-known method used in statistic to determine correlation between two or more variables, in machine learning for data classification and for prediction. This paper deals with data classification using a composition of two techniques of classification such as mesh generation, followed by the Hough transform method, in order to define a linear regression on the dataset. The purpose is to use analytical straight lines as a regression technique. The proposed method is a technique of data reduction through a mesh generation, which creates a virtual grid where each continuous point is localized in a cell corresponding to a pixel in the virtual grid. The standard Hough transform method establishes a relation between an image space and a parameter space through the definition of a sine function. The values of variables are represented by the coordinates of continuous points in the image space. The standard Hough transform is applied to each cell. The coordinates parameters with the high number of votes superior to a threshold that appears in the parameter space or the accumulator correspond to the parameters of analytical straight lines. Straight line recognition achieves linear regression and establishes correlation between the initial variables. Our analysis presents difference between classical linear regression and the proposed alternative which gives more visibility between data relation and accept a level of error in the dataset.

Data Classification, Mesh Generation, Hough Transform