Dissertation Title: FORECASTS USING HYBRID METHODS: QUANTITATIVE AND INTELLIGENT
Authors: Malvina XABAFTI
-
Institution: University of Tirana, Faculty of Economics, Department of Statistics and Applied Informatics
-
Field of study: Information Systems Economy
-
Publication date: 11.06.2026
-
The dissertation is published in Albanian.
© Copyright: Malvina XABAFTI
Published by the University of Tirana
Based on legal acts, regulations and policies of the UT
👉 Click here to view the full dissertation (PDF)
Abstract
This dissertation deals with the prediction of hydroclimatic extremes through a hybrid framework that integrates statistical time series models with intelligent methods (neural networks), with the aim of increasing the accuracy and warning capability for critical episodes. The study focuses on two phenomena of importance for the Albanian context: droughts through the SPEI6 index in the Fier area and floods through cumulative rainfall P3d in the Shkodra area. Individual statistical (SARIMAX) and neural (ANN, LSTM) models are compared with hybrid architectures built according to the “base model + correction” principle, where the statistical component captures the main structure of the signal, while the neural component learns the nonlinear part through residual learning. The evaluation is carried out in two dimensions: regressive forecasting of the series and identification/classification of extreme events, using appropriate metrics for unbalanced classes and time-based testing to preserve chronological integrity. The results of the comparative analysis show that hybrid models are more balanced and effective than individual models because they simultaneously improve the accuracy of the regression forecast and the warning ability for identifying extreme events. Within each phenomenon, hybrids provide more complete performance by reducing the average error and increasing the quality of signaling for rare events, while the comparison between phenomena confirms that the advantage of hybridization is maintained even when the process dynamics change, making the hybrid approach a sustainable forecasting strategy for hydroclimatic extremes.
Field: Statistics and Applied Mathematics in Economics
Keywords: Hybrid method, SARIMAX, ANN, LSTM, residual learning, drought, SPEI6, flood, P3d, extreme events;
Abstract
This dissertation addresses the forecasting of hydroclimatic extremes through a hybrid framework that integrates statistical time-series models with intelligent methods (neural networks), aiming to improve both predictive accuracy and early-warning capability for critical episodes. The study focuses on two phenomena of high relevance in the Albanian context: droughts, represented by the SPEI6 index in the Fier region, and flood-related extremes, represented by cumulative 3-day precipitation (P3d) in the Shkodër region. Standalone statistical (SARIMAX) and neural models (ANN, LSTM) are benchmarked against hybrid architectures designed under a "base model + corrector" principle, where the statistical component captures the dominant structure of the signal, while the neural component learns the remaining nonlinear dynamics via residual learning. Evaluation is performed along two dimensions: continuous (regression) forecasting of the target series and identification/classification of extreme episodes, using appropriate metrics for unbalanced classes and time-based testing to preserve chronological integrity. Comparative results show that hybrid models are more balanced and more effective than standalone models because they simultaneously improve regression forecasting accuracy and early-warning performance for extreme-event identification. Within each phenomenon, hybrids deliver more comprehensive performance by reducing average error and strengthening the quality of signaling for rare events, while cross-phenomenon comparisons confirm that the hybrid advantage remains robust even when process dynamics change, supporting hybridization as a stable forecasting strategy for hydroclimatic extremes.
Field of study: Applied Statistics and Mathematics in Economy
Keywords: Hybrid methods, SARIMAX, ANN, LSTM, residual learning, drought, SPEI6, floods, P3d, extreme events;