Basalt fiber-reinforced concrete (BFRC) mixed with fly ash, combined with advanced machine learning techniques, offers a practical, cost-effective, and less time-consuming alternative to traditional experimental methods. Conventional approaches to evaluating mechanical properties, such as compressive and splitting tensile strengths, typically require sophisticated equipment, meticulous sample preparation, and extended testing periods. These methods demand substantial financial resources, specialized labor, and considerable time for data collection and analysis. The integration of machine learning provides a transformative solution by enabling accurate prediction of concrete properties with minimal experimental data. The methods of data collection from literature and analysis were used and 121 records were collected from experimentally tested basalt fiber reinforced concrete samples measuring the compressive and splitting tensile strengths of the concrete. Eleven (11) critical factors have been considered as constituents of the studied concrete to predict the Fc-Compressive strength (MPa) and Fsp-Splitting tensile strength (MPa), which are the output parameters. The collected records were divided into training set (96 records = 80%) and validation set (25 records = 20%) following the requirements for data partitioning for sustainable machine learning application. Seven (7) selected machine learning techniques are applied in the prediction. Further, performance evaluation indices were used to compare the models' abilities and lastly, the Hoffman and Gardener's technique was used to evaluate the sensitivity of the parameters on the concrete strengths. At the end of the exercise, results were collated. In predicting the compressive strength (Fc), AdaBoost similarly excels, matching XGBoosting's validation performance with R2 of 0.98 and the same MAE values. This shows the effectiveness of boosting techniques for predictive modeling in concrete strength estimation. For splitting tensile strength (Fsp), AdaBoost also outperforms most models, achieving an R2 of 0.96 for training and validation phases. Its exceptionally low validation MAE of 0.124 MPa underscores its excellent generalization capabilities. Overall, XGBoosting and AdaBoost consistently demonstrate superior performance for both compressive and splitting tensile strength predictions, followed closely by KNN. These models benefit from advanced ensemble techniques that efficiently handle non-linear patterns and noise. SVR also performs admirably, whereas GEP and GMDHNN exhibit weaker predictive capabilities due to limitations in handling complex data dynamics. For the sensitivity analysis, the Hoffman and Gardener's method of sensitivity analysis proves instrumental in identifying key drivers of strength in fiber-reinforced concrete, guiding informed decision-making for material optimization and sustainable construction practices.