Articles
FORECAST OF FRUIT IMPACT BRUISING BY LINEAR, NON-LINEAR AND LOGISTIC REGRESSION MODELS
Article number
584_18
Pages
153 – 162
Language
English
Abstract
Fruit impact bruising is generally expressed by quantitative measurement of extent of damage or by limits and thresholds of the impact factors.
A different approach is the analysis of causes and factors connected to damage in order to express a global estimation of fruit susceptibility to bruising in terms of probability related to the impact event and the threshold of damage.
The authors have previously developed a drop damage index (DDI) for estimation of fruit damage potential.
This index was determined by a specific numerical procedure, where an estimate of quantitative damage expressed by a single value representing the probability of damage occurrence was determined from the drop height threshold for the fruit sample.
The model was based on multiple regression analysis considering the contribution of impact and different fruit variables on the mean damage.
Linear and non-linear multiple regression models were compared in terms of DDI calculation on apple, “Golden Delicious Klon B Lb”. The most reliable models in statistical terms were those with logarithmic algebraic operators (lne and log10) for drop height and fruit firmness.
Further, the application of a logistic regression model was analyzed in order to allow a direct estimation of fruit susceptibility in terms of impact bruising probability.
Logistic regression showed the highest prediction capability of all the models with drop height, firmness and fruit diameter as significant variables.
The percentage of damaged and undamaged fruits, correctly classified by the models was equal to 99.1%. By the means of logistic regression, impact bruising probability of 10% was found at 8 mm of drop height.
This value is equal to about one third of what was previously calculated by DDI, based on logarithmic models.
A different approach is the analysis of causes and factors connected to damage in order to express a global estimation of fruit susceptibility to bruising in terms of probability related to the impact event and the threshold of damage.
The authors have previously developed a drop damage index (DDI) for estimation of fruit damage potential.
This index was determined by a specific numerical procedure, where an estimate of quantitative damage expressed by a single value representing the probability of damage occurrence was determined from the drop height threshold for the fruit sample.
The model was based on multiple regression analysis considering the contribution of impact and different fruit variables on the mean damage.
Linear and non-linear multiple regression models were compared in terms of DDI calculation on apple, “Golden Delicious Klon B Lb”. The most reliable models in statistical terms were those with logarithmic algebraic operators (lne and log10) for drop height and fruit firmness.
Further, the application of a logistic regression model was analyzed in order to allow a direct estimation of fruit susceptibility in terms of impact bruising probability.
Logistic regression showed the highest prediction capability of all the models with drop height, firmness and fruit diameter as significant variables.
The percentage of damaged and undamaged fruits, correctly classified by the models was equal to 99.1%. By the means of logistic regression, impact bruising probability of 10% was found at 8 mm of drop height.
This value is equal to about one third of what was previously calculated by DDI, based on logarithmic models.
Publication
Authors
P. Menesatti, G. Paglia, S. Solaini, A. Zanella, R. Stainer
Keywords
bruising, impact, fruit, apples, statistical model, logistic regression
Online Articles (32)