A training dataset supplies the historical examples a supervised model uses to learn patterns for predicting unseen data. Each example pairs input attributes, also called independent variables, with a desired output: the target variable or label. section 1.5.1
Start with the prediction goal. For customer churn, a telecom record can contain revenue, service duration, average cost, and monthly usage alongside whether the customer churned. The attributes are inputs; the churn value is the label the model learns to predict. section 1.5.1
Table 1.3 Example of a structured dataset for a telecom operator showing multiple data attributes
|
ID
|
Revenue ($)
|
Duration of service (years)
|
Avg. cost
|
Monthly usage (days)
|
Churned (Y/N)
|
|---|---|---|---|---|---|
|
1001
|
100
|
1.1
|
0.10
|
10
|
Y
|
|
1002
|
200
|
4.1
|
0.09
|
25
|
N
|
|
1003
|
300
|
5.2
|
0.05
|
28
|
N
|
|
1004
|
200
|
0.9
|
0.25
|
11
|
Y
|
|
1005
|
100
|
0.5
|
0.45
|
12
|
Y
|
The telecom table makes the input–label pairing visible: each row contains customer attributes and a Churned (Y/N) value. section 1.5.1
The target determines the prediction task. Predicting a continuous value, such as weight from height, is regression; predicting a category, such as whether a transaction is fraudulent, is classification. Training quality matters because predictions depend on patterns learned from labeled historical data. section 1.5.1
Face identification
To train a neural network to identify people, pair face images with the corresponding names. During training, it learns distinguishing attributes from those examples; the detailed neural-network training process is separate from this dataset-level view. section 8.2.1
