Applying machine learning techniques for sea level rise forecasting in Axim: tackling missing data and outliers

Sea level forecasting is essential for effective coastal management and strategic planning, particularly in addressing the challenges posed by rising sea levels. Accurate predictions can guide informed decisions, such as resettlement strategies for vulnerable communities. However, the accuracy of time series models is often limited by the quality of data preprocessing and the underexploration of machine learning (ML) models for sea level forecasting, especially in regions with sparse datasets. This study addresses these gaps by conducting a comparative analysis of four ML models—K-Nearest Neighbors (KNN), Neural Network Auto-Regressive (NNAR), Extreme Gradient Boosting (XGBoost), and Facebook Prophet (FProphet)—against a classical non-seasonal ARIMA model, which serves as the baseline for comparison. Monthly sea level rise data for Axim, Ghana, spanning January 2000 to December 2018, was obtained from the Ghana Institute of Oceanography. Advanced preprocessing techniques, including Kalman smoothing for handling missing values and Winsorization for managing outliers, were employed to enhance data quality. Model performance was evaluated using mean squared error (MSE), mean absolute error (MAE), and root mean squared error (RMSE). Results show that NNAR outperformed all models (MSE=0.0041, MAE=0.0562, RMSE=0.0643), followed by FProphet (MSE=0.0250, MAE=0.1409, RMSE=0.1582) and ARIMA (MSE=0.3821, MAE=0.5585, RMSE=0.6182). KNN (MSE=0.8323, MAE=0.7549, RMSE=0.9123) outperformed the least-performing XGBoost (MSE=3.8573, MAE=1.9466, RMSE=1.9640). This study provides critical insights into robust forecasting approaches and strategies for managing the impacts of sea level rise in Axim.