Investigating the impact of the K-means clustering algorithm on SVM model performance in the task of predicting the exchange rate of the Syrian pound

Abstract

This research investigates the impact of the K-means clustering algorithm on the performance of a Support Vector Machine (SVM) model in predicting the exchange rate of the Syrian Pound against the US dollar. This study holds significant relevance given the substantial fluctuations observed in the Syrian Pound's exchange rate and its direct impact on the Syrian economy. The study utilizes daily time-series data for the exchange rate, spanning from the beginning of 2015 to mid-February 2024, comprising 3333 observations. SPSS26, RStudio, and Orange Data Mining were employed for data analysis and algorithm implementation.

The K-means algorithm was initially applied to cluster the data. Results indicated the algorithm's success in classifying the data into two statistically distinct clusters, suggesting underlying patterns within the exchange rate data. Subsequently, the SVM algorithm was applied to predict the exchange rate, first without utilizing K-means as a preprocessing step and then with its inclusion. The findings demonstrated a notable improvement in the SVM model's performance when K-means was employed as a preprocessing step. Specifically, the Mean Squared Error (MSE) decreased by 47.23%, the Root Mean Squared Error (RMSE) decreased by 27.28%, and the Mean Absolute Error (MAE) decreased by 18.21%. Conversely, the Mean Absolute Percentage Error (MAPE) increased by 31.81%. However, a substantial increase in the coefficient of determination (R²) by 52.32% indicates a significant improvement in the model's ability to explain the variance in the data. Based on these results, the study recommends incorporating K-means as a preprocessing step to enhance SVM performance in exchange rate prediction. Furthermore, it recommends future research encompassing longer time periods and employing advanced analytical techniques such as deep learning.