Abstract-For several services via the Internet, cloud computing offers a flexible and cost-effective solution. The newest computingprototypes for pooled computing resources, such asstorage, bandwidth, processing power, servers, applications, &services, areconsidered a significant IT change. We propose two approaches in this paper, The first research approach presents a securedata center dependent on a FODPSO algorithm. The suggested FODPSO method has multiple PSOs for particle swarmoptimization, in which each particle strives for an optimal solution for its own "survival," with the inherent advantage of havingmemory of previous choices. The first disadvantage that the conventional PSO has pointed out is this new architecture: apremature swarm convergence. The FODPSO discards swarms that converge prematurely towards a solution that may or maynot be best, as does conventional Darwinian (DPSO). In the second proposed approach, the present study thus uses aConvolutional Neural Network (CNN), which converges quicker, to look for the optimal CSN design, to recognize the humanactivity. The use of PSO for training aims at optimizing the findings of solution vectors on CNN, which in turn increases theprecision of classification to ensure that quality performance compared with state-of-the-art designs is achieved. The secondresearch approaches examine PSO-CNN algorithms and compare the performances of conventional machine-dependentalgorithms and deep learning methods. While the findings for CNN in HAR are positive, many factors are required to identifythe optimum CNN design. Any neural network focuses mostly on minimizing errors between goals and anticipated outputs.Cross-entropy in the event of CNN is performed via back-propagation and gradient descent. There are numerous factors to seteven a basic CNN. So, algorithms that discover and assess the CNN architecture in less time are important to find.Keywords—Cloud Computing, FODPSO, PSO-CNN.