Table 4 summarizes the performance analysis and evaluation of the 3 learning algorithms.

Table 4: Summary of Performance Analysis and Evaluation

LearningAlgorithm Prediction accuracy for regular usage pattern Processingpowerrequirements Speed Complexity
MI Low Low Fast Low
SLP Good Low Fast Average
MLP Good High Slow High


In this paper, we have presented the design and development of an intelligent interface reconfiguration engine that is context-aware. Widget reconfiguration is done dynamically without the need for modeling effort. Test results show that both the Single Layer Perceptron with Error Correction and Multi Layer Perceptron with Back Propagation can be used for context-aware reconfiguration of the mobile phone interface. However, the Single Layer Perceptron with Error Correction offers a practical yet effective solution for a resource-constrained mobile phone. It offers low computational overheads with reasonable prediction accuracy for the typical mobile phone user. Although competitive performance is offered by the MLP, a period of learning with existing data is required. Together with higher computational overheads, it may not be suitable as an on-the-fly approach. Future work would include investigating the effectiveness of approaches that include fuzzy logic engines and/or the Kohonen neural network as well as deploying the system on an actual mobile phone integrated to suitable wireless sensor device inputs.