The quality of freshly harvested herbs is heavily influenced by multiple factors, namely weather conditions, harvesting, transport, drying, storage, and many more. Our main goal here is to identify models that are able to predict spore contaminations on different types of herbs on the basis of these factors as well as to find optimal processing parameters, which shall lead to lower contaminations of herbs as well as lower costs for contamination prevention represents. The here presented workflow utilizes two different approaches, which in combination shall lead to a reliable contamination prediction and prevention mechanism. For the prediction part we learn ensembles of machine learning models using the processing parameters as features to predict the risk for spore contamination a priori of labor analysis data. Using tree-based modelling algorithms we already achieved a spore contamination prediction accuracy of 86.21% for the herb nettle. In Addition to that, we use descriptive statistics to provide information on the relevant parameters which could be responsible for the occurred contamination. Here we already achieve a p-value smaller than 0.01 for a few processing parameters. In the future we want to expand this workflow by improving the modelling process using different modelling algorithms. Additionally, we are working on an online life system, which combine these two methods, to not only present a farmer the information whether a contamination is probably, but also provide him the information which processing parameters lead to a contamination and how they should be affected to lower the risk.
|Titel in Übersetzung||Datengestützte Vorhersage der Verunreinigung von Kräutern und Ernteempfehlungen|
|Publikationsstatus||Veröffentlicht - 18 Sep. 2020|
- Data preprocessing
- Applied Statistics
- Contamination classification
- Machine Learning