An intelligent human activity recognition system is influenced to some extent by sensor placement. In this paper, the number, and placement positions, of wearable accelerometers have been investigated to determine their influence on a human activity recognition system. Given 17 possible human sensor placements, we developed a multi-stage and multi-swarm discrete particle swarm optimization algorithm to explore the optimal sensor combination for various required sensor amounts. Relevant experimentation involved 10 different human daily activities, achieving an average prediction accuracy for a 4-sensor optimal combination of 95.12% via support vector machine classifier. The number and corresponding placement of sensors required for activity recognition have also been provided in this paper.