A FUZZY BASED DIVIDE AND CONQUER ALGORITHM FOR FEATURE SELECTION IN KDD INTRUSION DETECTION DATASET

This paper provides a fuzzy logic based divide and conquer algorithm for feature selection and reduction among large feature set of KDD intrusion detection data set, since a reduced feature set will help to evolve better mining rules.This algorithm introduces a fuzzy idea of dividing the normal record by attacks records or vice-versa, and then considers the feature sets for every attack type separately. Actually, this algorithm is applied on KDD CUP 99 dataset having 37 attack types and selecting important feature among 41 feature of KDD dataset. The selected features are used in TANAGRA [11, 12] data mining tool to classify the dataset (i.e. KDD 99) for every attack vs. normal using various classification algorithms [5, 6]. The result for feature selection and classification shows a reduced set and maximized classification rate respectively.

Array data processing techniques applied to long-period shear waves at Fennoscandian seismograph stations

abstract Array data processing is applied to long-period records of S waves at a network of five Fennoscandian seismograph stations (Uppsala, Umeå, Nurmijärvi, Kongsberg, Copenhagen) with a maximum separation of 1300 km. Records of five earthquakes and one underground explosion are included in the study. The S motion is resolved into SH and SV, and after appropriate time shifts the individual traces are summed, both directly and after weighting. In general, high signal correlation exists among the different stations involved resulting in more accurate time readings, especially for records which have amplitudes that are too small to be read normally. S-wave station residuals correlate with the general crustal type under each station. In addition, the Fennoscandian shield may have a higher SH/SV velocity ratio than the adjacent tectonic area to the northwest.SV-to-P conversion at the base of the crust can seriously interfere with picking the onset of Sin normal record reading. The study demonstrates that, for epicentral distances beyond about 30°, existing networks of seismograph stations can be successfully used for array processing of long-period arrivals, especially the S arrivals.