Improving time series forecasting by applying the sliding window approach

Our primary research involves forecasting the IT job market, where we study trends, residuals, and seasonalities. In this study, we focus on the impact of the sliding window technique on forecasting models. The sliding window approach in the machine learning process is aimed at enhancing the accuracy of the forecasting models. It involves partitioning the continuous time series into subsets of consecutive and overlapping periods, which enables the models to track temporal characteristics effectively. The experiment is carried out on various algorithms and integrated with the sliding window. The technique allows flexibility for models to adapt to changes in the data dynamics, which greatly reduces the errors in forecasting. The study shows that sliding window methods are quite useful for building dependable and adaptive forecasting models. LSTM, ARIMA, SARIMA, and Holt's Model were used in this experiment with a dataset of 1 048 576 job rows with job-related information. Metrics such as MSE, RMSE, and MAE were used to test the models. LSTM was found to be the most efficient because of its capability to learn complicated patterns and long-term dependencies, and showed model improvement of 0.248 on MAE, 2.649 on MSE, and 0.162 on RMSE when the sliding window was applied.