Accurate classification of handheld object weight during different human motion is crucial for applications in health monitoring, injury prevention and exoskeleton systems. This study investigates the feasibility of using only a single forearm mounted inertial measurement unit (IMU) combined with AI algorithms to classify both movement types and the weights of handheld objects. A series of experiments with one subject were conducted to collect IMU data under various combinations of movements and object weights. Multiple feature extraction techniques including time, frequency, and time-frequency domains were applied, followed by classification using machine learning methods (SVM, KNN) and deep learning models (1D-CNN-LSTM, Wavelet-CNN). A genetic algorithm was used for optimal feature selection in machine learning pipelines, while open set classification capability was implemented using the Convolutional Prototype Network (CPN). Result shows that deep learning models, particularly 1D-CNN-LSTM method, outperform machine learning methods, achieving up to 94% classification accuracy. Moreover, the CPN model effectively rejected unknown movement patterns in open set scenarios. The proposed framework shows promising potential for wearable systems capable of intelligent workload classification in real world environments.