Forecasting involves learning trends and patterns in historical records and calculating the probability for an event to happen to make a statement about the events whose actual outcomes are yet to be observed. It became a critical tool in decision science applied across various domains, providing insights into future results to minimize uncertainty. Challenges persist while forecasting methods have evolved, including limited data, complexity, uncertainty, and representation. In this paper, we introduced the data framing method, which is a novel feature engineering that involves subtracting the present observation from all data in the window, centering the data around the current word. The method improves the traditional sliding window representation of time series. A comparative analysis between sliding window and data framing method in application to forecasting model using Artificial Neural Network. Experimental results demonstrate the superiority of the data framing method in terms of training convergence, validation loss fluctuation, and overall accuracy across diverse data against the sliding window method. The data framing method enhances the performance of ANNs.