This paper introduces a method for optimizing wearable sensor placement in warehouse weight-lifting tasks to enable accurate classification of lifted weights while minimizing hardware requirements. The approach employs a virtual human mesh model to identify the most informative sensor location by analyzing the maximum standard deviations of orientation readings. A Long Short-Term Memory (LSTM) neural network processes data from the optimally positioned sensor to classify loads into six weight categories. Key contributions include: first, reducing the number of sensors required by determining the most informative anatomical position and, and second, enabling accurate weight estimation without relying on muscle activity measurements. Experimental evaluation demonstrates a 95% probability of correctly classifying the lifted load. The proposed method allows for precise load weight estimation without requiring physiological data.