The calibration of internal combustion engines is a time-consuming and costly process that traditionally relies on extensive measurements and large datasets. This paper proposes an ANN-based method for estimating engine load from only four inputs: design power (crankshaft power computed from measured torque and rated speed), wheel power, torque, and engine rotational speed. The proposed model achieves superior performance, including approximately 99% accuracy in multiclass classification and strong regression metrics, by using a streamlined dataset collected through dynamometer and HiL setups. The ANN architecture uses multiple hidden layers with ReLU activation and a softmax output for classification. It is optimized using sparse categorical cross-entropy loss. Experimental results show that the model effectively captures nonlinear relationships between input parameters and engine load and outperforms conventional machine learning techniques. This approach significantly reduces calibration complexity and resource requirements while enabling real-time applications, such as predictive maintenance and anomaly detection. Furthermore, the framework is scalable and adaptable to other powertrain systems, including electric motors.