Human activity recognition based on feature selection in smart home using back-propagation algorithm.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Smart home environments generate vast amounts of sensor data.
• Accurate human activity recognition is essential for intelligent home automation and user assistance.
• Existing probabilistic models like Naïve Bayes and Hidden Markov Models have limitations in complex activity recognition tasks.

Purpose of the Study:

• To evaluate the effectiveness of the Back-propagation (BP) algorithm for training feedforward neural networks in human activity recognition.
• To investigate the impact of feature selection methods, specifically the inter-class distance method, on recognition performance.
• To compare the performance of the BP-trained neural network against Naïve Bayes (NB) and Hidden Markov Model (HMM) classifiers.

Main Methods:

• Utilized the Back-propagation (BP) algorithm to train a feedforward neural network.
• Applied the inter-class distance method for feature selection from motion sensor data.
• Evaluated human activity recognition performance using metrics like accuracy.
• Compared the BP neural network with NB and HMM classifiers.

Main Results:

• The choice of feature datasets significantly influences activity recognition accuracy.
• Suboptimal feature selection increases computational complexity and reduces recognition performance.
• The neural network trained with the BP algorithm demonstrated superior human activity recognition performance compared to NB and HMM.

Conclusions:

• The Back-propagation algorithm offers a robust approach for human activity recognition in smart homes.
• Effective feature selection is critical for optimizing both accuracy and computational efficiency.
• BP-trained neural networks provide a more effective solution than traditional probabilistic models for this task.