In the context of globalization and intense market competition, mergers and acquisitions (M&A) have become a common strategy for enterprises to expand and transform. However, goodwill issues in M&A are increasingly concerning. Traditional goodwill impairment prediction models face drawbacks like reliance on precise predictions, large datasets, complexity, poor interpretability, high computational costs, and data acquisition difficulties. This paper proposes a prediction model based on an improved Mamba algorithm. By processing financial data from listed companies in the CSMAR database, the model constructs lagged and rolling statistical features to reflect performance trends. The Mamba algorithm dynamically adjusts input parameters through a Selective State-Space Model (SSM), capturing dependencies in long-sequence data and improving prediction accuracy and timeliness. Experimental results show the Mamba algorithm excels in handling long-sequence data, offering valuable guidance for financial management and risk control.