Machine learning methods enhanced with data preprocessing steps in drought prediction

Drought is a complex natural phenomenon that causes numerous adverse effects worldwide, including the depletion of water resources, reduced agricultural productivity, and disruption of ecosystem balance. This study aims to develop advanced models for drought detection, offering a scientific approach to combating climate change. The in-depth analysis of drought seeks to enhance our understanding of this critical issue in the scientific community and contribute to the development of sustainable solutions. Drought detection models have been implemented by applying various class imbalance processing techniques and dimensionality reduction methods. The primary focus is to evaluate the impact of the Synthetic Minority Over-sampling Technique and Near Miss sampling on the performance of the drought prediction model. Additionally, Principal Component Analysis and Linear Discriminant Analysis are employed to reduce feature dimensions and enhance model efficiency. The dataset is preprocessed using the Synthetic Minority Over-sampling Technique and Near Miss methods to address class imbalances. Subsequently, Principal Component Analysis and Linear Discriminant Analysis are applied to reduce the feature space while preserving critical information. The effectiveness of the proposed methodologies is assessed by training machine learning algorithms in different combinations and developing a hybrid deep learning model on the preprocessed data. The results indicate that SMOTE-based sampling significantly improves model performance, particularly in terms of accuracy, precision, recall, and F1 score. Furthermore, the combination of dimensionality reduction techniques proves to be effective in enhancing the overall reliability of drought prediction models. In addition to evaluating algorithms with different combinations and preprocessing techniques, this study distinguishes itself from the existing literature by exploring the hybrid use of these algorithms. The study stands out in the field of artificial intelligence by addressing real-world problems through an effective methodological approach. This research demonstrates that class imbalance processing and dimensionality reduction techniques contribute to the development of robust drought detection models. The findings emphasize the importance of selecting appropriate preprocessing steps to address class imbalances and strategically utilizing dimensionality reduction techniques to enhance model performance.