Classifier: Income predictor
An alternative classifier assignment can be constructed from another data set out of the UCI Machine-Learning Repository. The “Adult Data Set” can be used to predict whether someone’s income will be greater than $50,000 or not. The process is the same as for classifying cancer as benign or malignant.
In this case, there are three information fields that can be ignored. They are labeled “Fnlwgt” (field 3), “Native-country” (field 14), and “Education” (field 4). The last one is ignored because that value is captured in the adjacent field “Education-num.”
The other difference in this data set is that some fields have discrete attributes. Calculating averages for attributes such as “Hours per week” is simple: just add up all the values and divide by the count of values. The discrete attributes are a little more interesting. Imagine that we have 10 records, all of which are examples of the >50K examples. If the “Relationship” attribute for 2 of these 10 records is “Wife,” 3 is “Own-child,” 2 is “Husband,” 1 is “Not-in-family,” 1 is “Other-relative,” and 1 is “Unmarried,” then the “Relationship” attribute in the >50K model would be as follows:
Relationship Wife: 0.2
Own-child: 0.3
Husband: 0.2
Not-in-family: 0.1
Other-relative: 0.1
Unmarried: 0.1
Follow the six steps outlined for the cancer classifier.