New Method for Instance or Prototype Selection using Mutual Information in Time Series Prediction ## AbstractThe task of selecting an adequate subset of input vectors that are included in a training set when classifying, approximating or predicting an output is a relevant task that, if accomplished correctly, can provide storage and computational savings and im- prove the accuracy of the results. This problem can be tackled from different perspectives: outlier detection or instance selection of noise-free data. The concept of outlier was firstly introduced by Grubbs in [1] as: ”...an outlying observation, or outlier, is one that appears to deviate markedly from other members of the sample in which it occurs...”. This concept was later extended by Barnett and Lewis [2] as: ”... An observation (or subset of ob- servations) which appears to be inconsistent with the remainder of that set of data...”. Many other notations (novelty detection, anomaly detection, noise detection, deviation detection, excep- tion mining) have been used for this problem [3].
[Edit] |