This paper presents a multi-factor adaptive statistical arbitrage model that extends traditional cointegration-based strategies by incorporating multiple factors beyond price data for portfolio selection. The authors explore three methodologies: K-means clustering, graphical lasso, and a hybrid combination of both, to identify candidate portfolios that are likely to be cointegrated. The results demonstrate that clustering alone yields better candidate portfolios than using graphical lasso on the entire equity pool, and the hybrid approach further improves performance. The study also examines an adaptive approach that re-computes portfolios during the trading period to account for changing relationships, but finds that this re-learning does not enhance results. The strategy successfully passes the statistical arbitrage test at a statistically significant level, and the findings are validated on an independent dataset, confirming the robustness of the proposed model. Overall, the paper contributes a novel framework for statistical arbitrage that leverages multi-factor information and clustering techniques to enhance portfolio selection and trading performance.