The overloading of heavy vehicles poses significant risks to road safety and infrastructure. Accurate and real-time payload monitoring is critical for mitigating these risks. In this study, we aim to develop a cost-effective and simplified method for payload estimation in heavy-duty trucks using strain-gage measurements on semi-elliptic leaf springs. A complete system was constructed, incorporating Wheatstone bridge circuitry and signal amplification, followed by experimental calibration and validation under controlled loading conditions. Additionally, theoretical strain-to-load relationships were derived on the basis of classical beam theory and evaluated against experimental results. To overcome the limitations of theoretical estimation, such as nonlinearities and structural uncertainties, a machine learning model using artificial neural networks (ANNs) was developed. A total of 255 data samples were collected from 17 repeated trials across a load range of 0 to 1300 kgf. These were randomly divided into 70% training, 15% validation, and 15% testing subsets. The trained ANN model achieved superior accuracy compared with traditional theoretical approaches. Results indicate that the ANN-based system provides more reliable load predictions than the theoretical strain–load estimation based on classical beam theory, as evidenced by the reduced prediction error (7% compared with 22%), and, owing to its low computational requirement, is suitable for deployment on edge devices such as ESP32 for real-world vehicular applications. The proposed approach offers strong potential for integration into IoT-enabled intelligent transportation systems.