Time series forecasting for uni- variant data using hybrid GA-OLSTM model and performance evaluations.

Area of Science:

• Environmental Science
• Data Science
• Artificial Intelligence

Background:

• Accurate time series forecasting of rainfall is crucial for water resource management and climate change adaptation.
• Traditional models often struggle with the complex, non-linear dynamics inherent in rainfall data.

Purpose of the Study:

• To develop and evaluate a hybrid Genetic Algorithm-Optimized Long-Short Term Memory (GA-OLSTM) model for uni-variant rainfall time series forecasting.
• To compare the performance of the GA-OLSTM model against a standard LSTM model.

Main Methods:

• A hybrid approach integrating a Genetic Algorithm (GA) with an Optimized Long-Short Term Memory (OLSTM) network was employed.
• The GA was utilized to optimize key parameters, including window size and the number of LSTM units, across various epochs (5, 10, 15, 20).
• Model efficiency was evaluated using Mean Square Error (MSE), Root Mean Square Error (RMSE), Cosine Similarity (CS), and correlation coefficient (r).

Main Results:

• The GA successfully identified optimal window sizes and unit numbers for different epochs, e.g., (12, 8) for 10 epochs.
• For 10 epochs, the GA-OLSTM model achieved superior performance metrics: MSE of 0.004, RMSE of 0.078, CS of 0.947, and r of 0.917.
• The standard LSTM model at 10 epochs yielded MSE of 0.006, RMSE of 0.078, CS of 0.910, and r of 0.858.

Conclusions:

• The Hybrid GA-OLSTM model demonstrates significantly enhanced accuracy and reliability in uni-variant rainfall forecasting compared to the standard LSTM model.
• The optimization of LSTM parameters via Genetic Algorithms is an effective strategy for improving time series prediction performance in meteorological applications.