Πρόβλεψη ζήτησης με χρήση μηχανικής μάθησης και βάσεων δεδομένων στην εφοδιαστική αλυσίδα
Demand forecasting using machine learning and data bases in supply chain
View/
Keywords
Μηχανική μάθηση ; Πρόβλεψης ζήτησης ; ΤαξινόμησηAbstract
The aim of this master's thesis is to analyze demand in the supply chain. More specifically, after reading this thesis, the reader will be able to understand the concept of demand, its determining factors, and how to make corresponding predictions through quantitative machine learning methods. The thesis assumes that the reader has no prior exposure to statistical methods and databases, and for this reason, it includes a detailed introduction to fundamental statistical concepts such as mean, variance, standard deviation, etc. Additionally, the thesis involves the development of a simple database for managing the products of a small electronics store.
Subsequently, after analyzing the basic concepts, statistical demand forecasting methods are presented, such as autoregression, the moving average, the combination of these two (ARMA), and the ARIMA method. The characteristics of these methods are thoroughly analyzed, including stationarity, seasonality, cyclicality, and so on.
Following this, demand analysis is carried out using machine learning models such as decision trees, Naive Bayes, k-nearest neighbors, linear regression, and logistic regression, for both classification and forecasting the next values of time series data. Finally, more modern models are applied for demand forecasting, such as neural networks, long short-term memory (LSTM) cells, and Facebook Prophet.
For the above analyses, the Python programming language is used. The classification data pertains to the likelihood of a customer repurchasing from a store (customer churn), based on various variables such as age, gender, the amount spent on their last purchase, and others. Demand forecasting is demonstrated through two examples: predicting the price of the Samsung S24 smartphone, and analyzing biannual sales of electronic device stores (Plaisio) from 2012 to 2024.