Continuous Attributes

• List for continuous attributes are kept in sorted order
• The candidate split points are the mid points between every two consecutive attribute values in training data
• For determining the split for an attribute for a node, two histograms are used :
  C_below stores the frequency of classes in which the attribute belongs for the records that are scanned
  C_above stores the frequency of classes in which the attribute belongs for the records yet to be scanned.
• C_below is initialized to zero and C_above is initialized with the class distribution for all the records for the node.
• C_below and C_above are evaluated and updated everey time when attribute record are read
• If winning split point found during scan, it's saved and C_below and C_above are deallocated before processing next attribute