Epilepsy is the common disease occurred due to the unusual electrical flow in the affected portion of the brain that causes seizure symptoms and hampers the normal behaviour of a healthy person. Researchers put their emphasis to develop powerful models to detect the particular types of seizure and presently machine learning algorithms are proved the best model to accurately detect the typical seizure. In this context, stochastic strategies are integrated with the conventional metaheuristic optimization algorithms to achieve global solution by improving exploration and exploitation mechanism in global search space. In this context, the conventional AOA is applicable for many real time applications but it suffers from major shortfalls of reaching towards the convergence point and attempts to fall to a local search space. An efficient version of AOA called as Levy Flight-Based Arithmetic Optimization Algorithm (LFAOA) is developed to improve the search capacity. The non-linear inputted signals are extracted with the most popular signal sample extraction technique called Discrete Wavelet Transform (DWT) and classified with widely used rule-based classifiers viz. Support Vector Machine (SVM), Least-Square SVM (LSSVM), Adoptive Neuro-fuzzy Inference System (ANFIS) etc. The stated technique DWT-LFAOA-LSSVM is tested with publicly available CHB-MIT epilepsy seizure dataset. In this experiment, obtained result is proved with Accuracy (ACC), Sensitivity (SN), Specificity (SP), Positive Predictive Value (PPV), Matthews Correlation Coefficient (MCC) and Area Under Curve (AUC) as 99.8%, 99.35%, 99.82%, 96.85%, 98% and 1 for DWT-LFAOA-LSSVM technique which supersedes with DWT-LFAOA-SVM and DWT-LFAOA-ANFIS. This proposed method can be tested with other publicly available datasets.