Sanal yemek verisi üzerinde talep tahmin sistemi tasarımı ve başarım değerlendirmesi

Demand prediction is a critical component in optimizing supply chain management and enhancing operational efficiency in various industries. By predicting demand in advance, manufacturers can plan their production, make informed decisions regarding distribution and storage options, and ensure that the amount of perishable goods meets real demand. However, demand prediction in the fresh food industry presents significant challenges due to the short shelf life and high demand variability of these products. Additionally, external factors such as weather conditions and seasonal changes make predicting supply and demand for fresh food even more complex. These complexities necessitate sophisticated prediction models that can adapt to rapid changes and incorporate real-time data. Implementing such models can help mitigate waste, decrease overproduction, optimize inventory levels, and ultimately improve customer satisfaction by ensuring fresh products are consistently available. This thesis explores advanced methodologies and algorithms seven for accurately predicting demand, particularly focusing on online food market data. For this purpose, we used two food related datasets; Online Market Dataset and a dataset from Kaggle called Food Demand Forecasting. As models we used Linear Regression, Decision Tree Regressor, Random Forest Regressor, Gradient Boosting Regressor, eXtreme Gradient Boosting, Multi-layer Perceptron Regressor, Light Gradient Boosting Machine, and Nonlinear Autoregressive Exogenous Model. The first seven models were used as a benchmark for the time series model, Nonlinear Autoregressive Exogenous Model. The performance metrics used to evaluate the models' performance are Root Mean Squared Error, R-squared, Mean Squared Error, and Mean Absolute Error. Among the seven machine learning models, for the Online Market Dataset, the Decision Tree Regressor and Random Forest Regressor exhibit the best performance. In addition, for the Food Demand Forecasting dataset, the Light Gradient Boosting Machine model exhibits the best performance, having the lowest Root Mean Squared Error (138,45) and highest R-squared (0,87), as well as the lowest Mean Squared Error (19.169,66) and a low Mean Absolute Error (75,51). Lastly, compared to other models, the Nonlinear Autoregressive Exogenous Model shows moderate performance. It performs significantly better than the Linear Regression model. However, the Nonlinear Autoregressive Exogenous Model is outperformed by other models. Keywords: Food Market, Demand Prediction, Nonlinear Autoregressive Exogenous Model