Domain knowledge is understanding of the real-world business context behind data. Machine learning can find patterns in historical data, but interpreting those patterns also requires knowing what would make sense in practice. section 1.4 section 4.3.1
Use domain knowledge when turning a measured relationship into a decision. Lift helps assess an association rule, but its value alone does not establish that the rule has a credible business interpretation. Check whether the proposed relationship fits how customers or processes actually behave before acting on it. section 4.3.1
Check the rule, not just the metric
A rule involving eggs and bread as substitutes for cheese may receive a metric-based interpretation yet fail a real-world check. A rule linking purchases of milk and eggs to bread is more plausible for some customers. Business logic can support or reject either conclusion before a team uses it. section 4.3.1
Table 4.2 Dataset to understand the concept of support, confidence, lift, and conviction
|
Invoice Number
|
Milk
|
Eggs
|
Bread
|
Cheese
|
|---|---|---|---|---|
|
1001
|
1
|
1
|
1
|
0
|
|
1002
|
0
|
1
|
1
|
1
|
|
1003
|
1
|
1
|
1
|
0
|
|
1004
|
0
|
1
|
0
|
1
|
|
1005
|
0
|
1
|
1
|
0
|
The five-invoice transaction table shows the milk, eggs, bread, and cheese purchases from which the example association rules are assessed. section 4.3.1
The goal is an actionable insight, not merely a detected pattern. Relating observations produces knowledge; identifying a significant point that a business team can act on takes interpretation. For example, a manufacturing team could respond to evidence that one mold produces substantially more defects than another. section 1.2.1
