Falls are the leading cause of injury and death in people over 65. Timely detection and warning of the fall risks of humans, especially the elderly, while performing daily living activities are vitally necessary. Therefore, this paper proposes a Dense-Inception Neural Network (DINN) to classify falls among 11 human activities based on micro-Doppler signatures. The network's hyper-parameters are analyzed and fine-tuned through experiments with the simulated dataset from Simhumalator software to choose the most optimal network model. As a result, the proposed model with 24 filters achieves a good balance between prediction time and classification accuracy performance. Moreover, the proposed model's results remarkably outperform when compared with four other networks with the same input dataset due to the dense-inception structure.