Human Activity Recognition (HAR) is a critical task in computer vision, based on classifying human actions based on diverse sensor inputs or sequential video frames. This proposed work implements an innovative approach to HAR, employing a Convolutional Long Short-Term Memory (ConvLSTM) network that combines convolutional neural networks (CNNs) and long short-term memory networks (LSTMs). The model architecture, detailed in furthur sections, comprises multiple ConvLSTM layers with features like batch normalization, max pooling, and dropout layers, forming a sophisticated deep learning framework tailored for HAR. The model emphasizes the temporal and spatial details and intricacies of human actions. Recurrent dropout is applied to enhance the network’s generalization capabilities. The incorporation of ConvLSTM layers with increasing filters, along with batch normalization, aims to capture hierarchical features in sequential video frames. Utilizing a 3D max pooling strategy and time-distributed dropout, the model emphasizes robust feature extraction while preventing overfitting. The final layers include flattening and a dense layer with a softmax activation, enabling the model to predict human activity classes from a predefined list. The proposed model has an accuracy of 93.12%, with a loss of 0.003923. This comprehensive ConvLSTM architecture is designed to exploit the synergies between CNNs and LSTMs, enhancing the understanding and prediction of human activities in sequential video frames. The proposed work endeavors to advance the current state-of-the-art in HAR, showcasing the adaptability and complexity of the ConvLSTM model through its detailed summary and incorporation of specific data dimensions and layer configurations.