1 results
Search Results
Now showing 1 - 1 of 1
Master Thesis İhracat Miktarlarının Gelişmiş Zaman Serisi Tahmini İçin Transformatör Modellerinden Yararlanma(2024) Coşkun, Çağrı; Yıldız, Beytullah; Yazıcı, AliForecasting export amounts is crucial for small and medium-sized enterprises (SMEs) to remain competitive in global markets. Traditional machine learning methods often struggle with the complexities of multiple multivariate time-series analysis, where financial data is recorded annually for each company, showing irregular fluctuations and long-term dependencies. Address these challenges, we introduce a Transformer based approach for forecasting export amounts using annually repeated financial data. The Transformer model, with its advanced attention mechanisms, outperformed Random Forest and Long Short-Term Memory (LSTM) models on our dataset, which spans nine years for each enterprise. When the number of time points in the dataset was reduced, the Transformer model exhibited a significant drop in performance. However, its performance increased notably with the use of an extended time series, clearly showing that successful and impactful results require sufficiently long, feature rich time series, enhanced by effective feature engineering. These findings indicate that Transformer models can significantly improve the accuracy of forecasting complex time series based on financial data and offer valuable insights for SMEs and policymakers.
