Feature Selection for Supervised Learning and Compression

Supervised feature selection aims to find the signals that best predict a target variable. Typical approaches use measures of correlation or similarity, as seen in filter methods, predictive power learned models, wrapper methods. In both approaches, selected features often have high entropies and are not suitable for compression. This is particular drawback automotive domain where fast communication archival vehicle telemetry data increasingly important, especially with technologies such V2V V2X (vehicle-to-vehicle vehicle-to-everything communication). paper select good performances compression by introducing compressibility factor into several existing approaches. Where appropriate, performance guarantees provided greedy searches based on monotonicity submodularity. Using language entropy, relationship between relevance, redundancy, discussed from perspective signal selection. The then demonstrated selecting Controller Area Network SVMs regression task, namely predicting fuel consumption, classification identifying Points Interest. We show while slightly lower when considered, significantly improved.