Optimizing copious activity type classes based on classification accuracy and entropy retention

摘要: Despite the advantages, big transport data are characterized by a considerable disadvantage as well. Personal and activity-travel information often lacking, making it necessary to deduce this with mining techniques. However, some studies predict many unique activity type classes (ATCs), while others merge multiple types into larger ATCs. This action enhances inference estimation, but destroys important information. Previous do not provide strong justification for practice. An objectively optimized set of ATCs, balancing model prediction accuracy preserving from original data, becomes essential. research developed classification methodology in which optimal ATCs was identified analyzing all possible ATC combinations. approach is practically impossible finite amount time e.g. US National Household Travel Survey (NHTS) 2009 set, comprises 36 (home excluded), since there would be 3.82•1030 combinations (an exponential increase). The aim paper optimize should grouped new class, sets or impractical simply calculate proposed method defines an optimization parameter U (based on retention) maximized iterative local search algorithm. NHTS determined. A comparison finds that optimum considerably better than expert opinion systems. Convergence confirmed large performance gains were found.