@misc{GhulamZada2022, type = {Master Thesis}, author = {Ghulam Zada, Taj Mohammad}, title = {Machine-Learning-basiertes Forecasting des Elektromobilit{\"a}tshochlaufs in Deutschland}, doi = {10.25933/opus4-2878}, url = {http://nbn-resolving.de/urn:nbn:de:kobv:522-opus4-28784}, school = {Technische Hochschule Brandenburg}, year = {2022}, abstract = {Forecasting is an important decision-making tool for businesses, organizations, and governments because it saves a lot of resources. The forecast of the market share of new BEV car registrations in Germany can help efficiently evaluate strategic and future-oriented decisions, e.g., the climate targets set by the federal government. Machine learning methods were used to implement forecasting. Two machine learning models were used to forecast of the market share of new BEV car registrations in Germany: SARIMA(X) and DeepAR model. Due to variety in input datasets, four univariate and multivariate time series analysis models were trained. Afterwards, all four trained models were compared with each other using the RMSE metric. The last six months of the dataset were used to evaluate the model quality. The result of model comparison shows that SARIMAX model needs the exogenous variables to create a forecast of the multivariate data sets. However, the DeepAR model does not require any exogenous variables to generate forecast. Furthermore, if the input dataset contains at least 300 observations, then the implementation of DeepAR model is much more advantageous than SARIMA(X) model. DeepAR model not only supports the univariate and multivariate datasets, but also creates a global model that generates a forecast per variable. This is a great advantage over the SARIMA(X) model, which is not capable of this.}, language = {de} }