Imputing trip purposes for long-distance travel

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

“…This deficiency outweighs the fact that this scheme retains slightly more information than the optimal scheme. The scheme based on [45] performs similarly. The ones inspired by [44] and [30] perform worse than the 415 optimal scheme on both the classification accuracy and information retention.…”

“…This section discusses the combinatorial challenge of grouping or aggregating activity types into new classes. For example, in the set of ATCs [ [1], [2], [3], [4], [5], [6], [7], [8], [9], [10]], activity types 3 and 6 may be merged into a new class as such: [[1], [2], [3,6], [4], [5], [7], [8], [9], [10]]. The number of possible ATCs grows exponentially with the number of distinct activity types: n. This is the result of all the permutations of activity types across possible groups and the different combinations of possible group sizes.…”

“…The schemes inspired by [4], [42] and [43] have similar of better classification accuracies compared to the optimal scheme, however these lost a major portion of their information content as a consequence.…”

“…In order to overcome this shortcoming, behavioral data mining techniques 10 are frequently used to infer activity types (sometimes otherwise denoted as trip or travel purposes, activity classes, activity categories or activity encoding) from behavioral attributes, such as temporal attributes and spatial information (e.g. [3,4,5]). However, in these researches different classifications of activity types exist.…”

“…However, for the US National Household Travel Survey (NHTS) 2009 data set [8] which comprises 36 ATCs (home activity excluded, see section 3), calculating classifiers for all possible grouping combinations is impossible since the increase in distinct combinations is exponential. In other words, a large number of initial activity types which are considered for aggrega- 45 tion will result in an extremely large set of grouping combinations that needs to be processed as shown in Table 1. Subsection 4.1 will discuss how the number of possible ATCs in Table 1 can be calculated.…”