Recognition models often experience a drop in performance when applied to new users or when users’ physical or behavioral states change, which typically requires retraining with more labeled data. Additionally, challenges arise in search domains due to object interactions and the difficulty in distinguishing similar actions performed by different individuals. To overcome these limitations, a hybrid algorithm for improving recognition system in human activities was developed. This solution combines the capabilities of Convolutional Neural Networks (CNN) and Fuzzy Logic to leverage both deep learning and rule-based reasoning. The algorithm was built using a dynamic software development methodology, an agile framework well-suited for deep learning projects. The datasets were trained and tested using Google Colab using Python, TensorFlow, and Keras libraries. This hybrid model improved the accuracy and efficiency of human activity classification and was able to manage vague or uncertain data within the data warehouse. It also accurately identified specific actions within sequences of video frames. The model achieved an F1-score of 96.83%, a recall of 97.14%, and precision of 114.2% and an accuracy of 97.14%, marking a significant improvement over previously existing algorithms.