The increasing popularity of online job vacancies and machine learning methods has raised questions about their combination to enhance our understanding of labour markets and algorithms. However, the lack of comparable studies necessitates further investigation. This research aims to explore the effectiveness of Random Forest Regressor (RFR) and Support Vector Regressor (SVR) machine learning models in predicting online job vacancies compared to the auto-regressive ARIMA method. To answer this question, detailed sub-questions are posed in relation to the sub-samples of the main data provided by Birch Consultants, an external partner originally obtained by Jobdigger. Drawing upon previous research on time-series accuracy, this study combines various approaches to benefit society and the external partner. Using the walk-forward validation method, with a 91-day expanding window, it provides precise answers to the sub-questions. Findings suggest that RFR is suitable for forecasting larger samples, while SVR is preferred due to its capability to predict small series despite relatively small scoring benefits and computational costs. Both machine learning models outperform the baseline ARIMA model in capturing complex time-series. Further research should focus on exploring advanced auto-regressive, deep learning, and hybrid models for future investigations.