Browsing by Author "Rossouw, Christopher Guillaume"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    An investigation into table grape risk factors that affect quality along the export supply chain
    (Stellenbosch : Stellenbosch University, 2022-12) Rossouw, Christopher Guillaume; Goedhals-Gerber, Leila Louise; Mantadelis, Theofrastos; Stellenbosch University. Faculty of Economic and Management Sciences. Dept. of Logistics.
    ENGLISH SUMMARY: Table grapes are a highly perishable product, where a large proportion of grapes produced for export to Europe arrive in a substandard condition. Fruit in this condition requires repacking to remove the rotten food parts, or in extreme cases, the entire shipment is dumped resulting in a total loss. Both outcomes’ result in a potential loss of income for stakeholders and food waste, which could be avoided if proper upstream intervention had been taken. This ongoing occurrence prompted the investigation into what the factors are that cause the poor arrival quality of table grapes. The study also applied machine learning techniques to predict the probable arrival scores (green, amber, and red) based on input variables gathered throughout the supply chain. The data analysed was obtained from five diverse secondary sources consisting of intake quality shed reports, arrival quality reports, logistical nominal data, recorder temperature data, and climate data. The eventual dataset consisted of 467 observations. The analysis process applied consisted of descriptive and inferential statistics to explain the relationship between the upstream variables and the downstream resultant quality scores as well as how the upstream variables interact with one another. The results from the preliminary analysis aided in feature selection for the model building process. Four classification models, consisting of Logistic Regression, k-Nearest Neighbours, Decision Trees, and Random Forests (RF), were trained, and evaluated. The RF classifier demonstrated the best cross-validation score on the training data and was retained for further evaluation. The RF classifier’s accuracy score was 0.63 for the unseen test set and performed best when predicting red class-labels but struggled on green and performed worst for amber class predictions. Variables that had the largest impact on the arrival quality scores consisted of the climactic variables two weeks prior to harvest, the specific variety and ˚Brix at harvest, the number of decayed berries found in the packhouse as well as the overall packhouse quality score, and the type of packaging used (either punnets or loose pack). The effect of the supply chain was also evaluated but did not have any effect for the 2020 season. The attributes of poor quality were also identified in relation to the most important variables, so that upstream mitigation strategies could be determined to reduce financial claims and food waste. The potential upside of accurate arrival quality predictions prior to shipping would allow for improved allocation decisions leading to profit maximisation through loss reduction and cost savings. From an environmental perspective, assured sound arrival quality would reduce end of chain food waste and would increase product shelf life for consumers.