Advanced Data Mining Techniques - Neural Networks

Advanced Data Mining Techniques - Neural Networks

1. Artificial Neural Networks

   • Supervised
     • Feed-Forward NN
       • Bacpropagation
   • Unsupervised
     • Kohonen NN
   • Process
     • Split data into training & testing set
     • NN will learn and build model based on training set
     • Assign random weight for input variables
     • Read in the input and the desired output
     • Compute actual output and error
     • Adjust weight as it learns
     • Evaluate network using the test set
   • Network Training
     • Iterative process
       • Stops when
         • Error sum converged to below a present acceptable level
         • Predetermined iterations/passes is reached
   • Strength
     • Appropriate for both supervised and unsupervised
     • Use when results more important
     • Works well with noisy data
     • Can process numeric and categorical data
     • Can be used for applications that require a time element in the data
   • Weakness
     • Lack explanation capabilities
     • Require input range from 0 to 1
     • May converge on an inferior solution