Time series predictive analysis based on hybridization of meta-heuristic algorithms

Abstract

This paper presents a comparative study which involved five hybrid meta-heuristic methods to predict the weather five days in advance. The identified meta-heuristic methods namely Moth-flame Optimization (MFO), Cuckoo Search algorithm (CSA), Artificial Bee Colony (ABC), Firefly Algorithm (FA) and Differential Evolution (DE) are individually hybridized with a well-known machine learning technique namely Least Squares Support Vector Machines (LS-SVM). For experimental purposes, a total of 6 independent inputs are considered which were collected based on daily weather data. The efficiency of the MFO-LSSVM, CSLSSVM, ABC-LSSVM, FA-LSSVM, and DE-LSSVM was quantitatively analyzed based on Theil's U and Root Mean Square Percentage Error. Overall, the experimental results demonstrate a good rival among the identified methods. However, the superiority goes to FA-LSSVM which was able to record lower error rates in prediction. The proposed prediction model could benefit many parties in continuity planning daily activities.