Naslov
Predicting Peach Fruit Ripeness Using Explainable Machine Learning

Autori
Ljubobratovic, Dejan ; Guoxiang, Zhang ; Brkic Bakaric, Marija ; Jemric, Tomislav ; Matetic, Maja

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
Proceedings of the 31st International DAAAM Symposium "Intelligent Manufacturing & Automation" / - : DAAAM International Vienna, 2020, 0717-0723

ISBN
978-3-902734-29-7

Skup
31st International DAAAM Virtual Symposium "Intelligent Manufacturing & Automation"

Mjesto i datum
Mostar, Bosna i Hercegovina ; online, 21.10.2020. - 24.10.2020

Vrsta sudjelovanja
Demonstracija

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
interpretability ; explainable machine learning ; predicting fruit ripeness ; peach impedance

Sažetak
Predicting fruit ripeness allows us to choose the optimal time to harvest. The parameter by which peach ripeness is commonly represented is its firmness. As traditional methods for determining firmness of peaches are destructive, this paper uses an alternative method for determining peach ripeness which is based on peach impedance, as recommended by the domain expert. The data set on which the data analysis is performed contains measurements obtained from a couple of hundred fruit measurements, which also include peach impedance. In our data analysis, we use one of the high accuracy machine learning models, which are called black box models and which are characterized by low interpretability. The paper presents the results of applying a black box type machine learning method, as well as methods for interpreting black box models which facilitate understanding of the model behavior for domain experts, i.e. Variable importance, Tree Surrogate, Local Interpretable Model- Agnostic Explanations and Break Down.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo, Poljoprivreda (agronomija), Informacijske i komunikacijske znanosti

POVEZANOST RADA