Developed and integrated a novel ARAR forecasting model for time series analysis within the sktime/sktime repository, focusing on enhancing support for long-memory processes. The approach combined memory shortening through an adaptive autoregressive filter with subset AR modeling using Yule-Walker equations, resulting in a two-stage workflow that improves forecasting efficiency. Collaborated closely with other contributors to ensure code quality and comprehensive documentation of the model architecture. Leveraged Python and data science techniques to deliver this feature, emphasizing maintainability and alignment with project performance goals. The work strengthened the repository’s forecasting capabilities, particularly for complex time series data requiring advanced analytical methods.