DIMENSIONALITY REDUCTION BY INTRINSIC DIMENSION ESTIMATION USING BOX COUNTING METHOD AND CORRELATION DIMENSION METHOD

Dimensionality reduction is an essential phase in data mining. Intrinsic dimension is used to determine the attributes which covers the entire data set. Various techniques are involved in calculating intrinsic dimension. Here the intrinsic dimension for diabetes medical dataset is to be done through box-counting dimension and correlation dimension. Analysis shows that correlation dimension is more efficient than box counting method in terms of size of the dataset. User defined distance for calculating fractal dimension reduces the reliability of correlation dimension method, so log-log pairs of a data set is used in the correlation dimension. The sample size and number of redundant variables influence the computation of correlation dimension. Implementation of box counting and correlation dimension for diabetes data sets confirm the effectiveness of intrinsic dimension estimation with log-log pairs plot.