Enhanced multi-layer perceptron for CO2 emission prediction with worst moth disrupted moth fly optimization (WMFO)

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Population stagnation and low diversity are common issues in optimization algorithms like Moth Fluctuation Optimization (MFO).
• These limitations can hinder the algorithm's ability to find optimal solutions, leading to suboptimal performance in complex problems.

Purpose of the Study:

• To introduce the Worst Moth Fluctuation Optimization (WMFO) strategy to enhance the MFO algorithm.
• To address challenges of population stagnation and low diversity in MFO, improving global search capabilities.
• To apply the enhanced algorithm for accurate carbon emission prediction.

Main Methods:

• Development of the Worst Moth Fluctuation Optimization (WMFO) strategy as an enhancement to the Moth Fluctuation Optimization (MFO) algorithm.
• Evaluation of WMFO on CEC15 benchmark test functions, comparing its performance against MFO and other state-of-the-art algorithms using Friedman and Wilcoxon tests.
• Introduction of a hybrid model, WMFO-MLP, combining WMFO with a Multi-Layer Perceptron (MLP) for parameter tuning in carbon emission prediction.

Main Results:

• WMFO demonstrated a remarkable efficiency of 66.6% and outperformed MFO on CEC15 benchmark functions.
• Statistical tests (Friedman and Wilcoxon) confirmed WMFO's superiority over existing advanced algorithms.
• The hybrid WMFO-MLP model achieved an outstanding total accuracy of 97.8% in carbon emission prediction, surpassing alternative methods in precision, reliability, and efficiency.
• Feature importance analysis highlighted Oil Efficiency (40%) and Economic Growth (26.5%) as key predictors for carbon emission.

Conclusions:

• The WMFO strategy effectively overcomes population stagnation and low diversity issues in MFO, enhancing global search capabilities.
• The WMFO-MLP hybrid model offers significant advancements in optimization and predictive modeling, particularly for carbon emission prediction.
• The findings suggest practical applications of WMFO-MLP in environmental and economic forecasting, leveraging key factors like energy efficiency and economic indicators.