Optimizing Machine Learning Architectures for Time Series Forecasting: A Hybrid rvGA-eNM Approach

This study introduces a hybrid rvGA-eNM (real-valued Genetic Algorithm with enhanced Nelder-Mead) optimization approach for time series forecasting, specifically designed to address data scarcity and computational efficiency challenges in operational environments. Unlike contemporary hybrid algorithms that prioritize accuracy through increased complexity, rvGA-eNM employs adaptive algorithm orchestration that explicitly separates global exploration and local exploitation phases through convergence-based transition mechanisms. Multi-domain validation across Indonesian crude oil prices (156 monthly observations), Gorontalo regional electricity consumption (26 annual observations), and Albania GDP (125 quarterly observations) demonstrates robust forecasting performance with MAPE values ranging from 3.37% to 6.33% and convergence times between 0.93 and 3.31 seconds. Comparative benchmarking against state-of-the-art hybrid algorithms reveals substantial computational advantages: rvGA-eNM achieves comparable accuracy with 72× faster computation than deep learning hybrids and converges within 100 iterations compared to 200-500 iterations for contemporary methods. The algorithm exhibits exceptional small-sample robustness, maintaining reliable forecasts with as few as 26 observations—a critical capability for emerging markets and data-constrained applications. Cross-domain consistency, evidenced by narrow MAPE variance (2.96 percentage points) across fundamentally different forecasting contexts, suggests genuine algorithmic generalizability without domain-specific customization requirements. This research contributes Algorithm Orchestration Theory, formalizing how complementary algorithm capabilities combine synergistically through adaptive phase transitions. The findings challenge conventional assumptions about minimum data requirements in forecasting and demonstrate that computational efficiency deserves elevation as a primary objective alongside accuracy. The hybrid rvGA-eNM offers practitioners a practical, efficient solution for diverse operational forecasting applications, particularly valuable in resource-constrained environments where sophisticated forecasting methods have traditionally been inaccessible.