Much research has been done to detect human emotion using various computational methods, such as physiological measures and facial expression recognition. These methods are effective but they could be expensive or intrusive as special setups of equipments are needed. Some researchers have utilized non-intrusive methods by using mouse or keyboard analyses and presented comparable effectiveness in detecting human emotion. This paper investigates how both keyboard and mouse features can be combined to detect human stress, particularly cognitive stress induced by time pressure and mental arithmetic problems. The results show that the complexity of the mental arithmetic problem and time pressure affect user behaviour, mouse behaviour and keyboard behaviour significantly. This indicates that automatic analysis of human stress from keyboard and mouse input is potentially useful for providing adaptation in interactive system such as an e-learning system.