Waterflooding persists as the primary secondary oil recovery method, although its efficiency evaluation is often hindered by limitations from reservoir heterogeneities alongside non-uniform sweep efficiency and uncertainty in predicting well-to-well connectivity. To improve the evaluation of waterflooding, this research develops and validates nine advanced machine learning models, covering key types such as neural networks, linear models, tree-based models, ensemble learning, and support vector machines. This research utilized a dataset of 6592 observations collected across four wells over 2000 days to train and evaluate these machine learning models for fluid production rate forecasting and injector-producer connectivity assessments. The most successful model was AdaBoost since it delivered both a minimum root mean squared error (RMSE) of 0.004 and a coefficient of determination (R²) of 0.99, surpassing traditional methods. Furthermore, SHapley Additive Explanations (SHAP) analysis provided interpretability by quantifying injection wells' relative influence on production rates, facilitating the construction of a connectivity-driven contour map. Traditional evaluation methods of empirical and analytical approaches, as well as capacitance-resistance models and numerical simulators, function under restrictive assumptions, extensive computational requirements, and significant data demands. This study is novel in uniting multiple ML algorithms with SHAP interpretability and a connectivity-driven contour map for injector-producer dynamics, providing a practical, real-time workflow for field-scale waterflood optimization. The results of this study demonstrate that ML models surpass traditional methods in terms of predictive accuracy, computer processing speed, and scalability when applied for real-time waterflood optimization. This study illustrates the transformative potential of ML in reservoir engineering by enabling data-driven decision-making and improving the injection optimization process.