Accurately forecasting solar activity is critical for both climate modeling and space weather prediction. Traditional methods often rely solely on historical sunspot records, which may not fully reflect the complex physical processes driving solar variability. In this study, we propose a multivariate time series forecasting framework that integrates cosmogenic isotope data specifically Carbon-14 ( \(^{14}\textrm{C}\) ) and Beryllium-10 ( \(^{10}\textrm{Be}\) ) alongside historical sunspot numbers to enhance predictive performance. A unified dataset spanning several centuries is constructed and used to train advanced machine learning models, including XGBoost and LSTM, capable of capturing nonlinear and long-term temporal dependencies. For comparison, classical univariate models such as ARIMA and SARIMA are also implemented using sunspot data alone. Experimental results demonstrate that incorporating isotopic proxies significantly improves forecasting accuracy and robustness, particularly for long-term solar cycle variations. This study highlights the effectiveness of combining indirect solar proxies with modern machine learning techniques for the multivariate prediction of solar activity.