im PublikationsserverTitelangaben
Buchwitz, Benjamin ; Falkenberg, Anne ; Küsters, Ulrich:
Time Series Event Forecasting in Consumer Electronic Markets using Random Forests.
In: Proceedings of the 2019 Pre-ICIS SIGDSA Symposium. -
München, 2019
Volltext
|
Link zum Volltext (externe URL): https://core.ac.uk/download/pdf/301391654.pdf |
Kurzfassung/Abstract
Consumers are pricesensitive and opportunistic about the place of purchase when buying electronic goods. However, services that advise customers on their purchase time decisions for those products are missing. Given the objective to provide a binary signal to customers to either wait or purchase immediately, classification algorithms are a direct methodological choice. Approaches like random forests allow for the derivation of a probability and class prediction but are usually not used in time series contexts. This is due to missing or time-invariant regressors and unclear prediction settings. We show how classification methods can be used to generate reliable predictions of price events and analyze if they are subject to common market dependencies. Pooling univariate random forests and enhancing them with multivariate features shows that our approach generates stable and valuable recommendations. Because dependency structures between products are transferable, multivariate forecasting increases accuracy and issues recommendations where univariate approaches fail.
Forschungsprojekte
Weitere Angaben
| Publikationsform: | Aufsatz in einem Buch |
|---|---|
| Schlagwörter: | Price Event Forecasting, Multivariate Time Series, Random Forest, E-Commerce |
| Sprache des Eintrags: | Englisch |
| Institutionen der Universität: | Wirtschaftswissenschaftliche Fakultät > Statistik > Lehrstuhl für Statistik und Quantitative Methoden der Wirtschaftswissenschaften |
| Begutachteter Aufsatz: | Ja |
| Titel an der KU entstanden: | Ja |
| KU.edoc-ID: | 23791 |
Letzte Änderung: 08. Dez 2021 20:58
URL zu dieser Anzeige: https://edoc.ku.de/id/eprint/23791/
