Predicting successful trading in the West Texas Intermediate crude oil cash market with machine learning nature-inspired swarm-based approaches

Abstract

The subject of predicting global crude oil prices is well recognized in academic circles. The notion of hybrid modeling suggests that the integration of several methodologies has the potential to optimize advantages while reducing limitations. Consequently, hybrid techniques are extensively used in contemporary research. In this paper, a novel decompose-ensemble prediction approach is proposed by integrating various optimization algorithms, namely biography-based optimization (BBO), backtracking search algorithm (BSA), teaching-learning-based algorithm (TLBO), cuckoo optimization algorithm (COA), multi-verse optimization (MVO), and multilayer perceptron (MLP). Furthermore, the aforementioned approaches, namely BBO-MLP, BSA-MLP, and TLBO-MLP, include the de-compose-ensemble technique into the individual artificial intelligence model in order to enhance the accuracy of predictions. In order to validate the findings, the forecast is conducted using the authoritative data on oil prices. This study will use three primary indicators, including EMA 20, EMA 60, EMA 100, ROC, and AUC assessments, to assess and evaluate the efficacy of the five methodologies under investigation. The below findings are derived from the conducted research: Based on the achieved AUC values of 0.9567 and 0.9429, it can be concluded that using a multi-verse optimization technique is considered the most suitable strategy for effectively handling the dataset pertaining to crude oil revenue. The next four approaches likewise have a significant AUC value, surpassing 0.8. The AUC values for the BBO-MLP, BSA-MLP, TLBO-MLP, and COA-MLP approaches were obtained as follows: (0.874 and 0.792) for training and testing stages, (0.809 and 0.792) for training and testing stages, (0.9353 and 0.9237) for training and testing stages, and (0.9092 and 0.8927) for training and testing stages, respectively. This model has the potential to contribute to the resolution of default probability and is very valuable to the credit card industry. Broadly speaking, this novel forecasting approach serves as a notable predictor of crude oil prices.