Data Selection in Data Mining
Data selection is the process of choosing the right type of data, data
source, and tools to collect data before starting a project.
It is an important step because the quality of your results depends on the
data you choose.
Main Goal of Data Selection
The goal is to select data that can properly answer the research question.
This depends on:
- The purpose of the study
- Previous research (literature)
- Availability of data
If data is chosen only based on low cost or convenience, it may reduce the
accuracy and
reliability of the results.
Common Issues in Data Selection
While selecting data, researchers should consider:
- Choosing the correct type and source of data
- Ensuring the data can answer the research question
- Collecting a representative sample
- Using proper tools/instruments for data collection
- Making sure the data and tools are compatible
Types of Data
There are two main types of data:
1. Quantitative Data
- Numerical data
- Example: temperature, height, voltage
- Used for calculations and analysis
2. Qualitative Data
- Non-numerical data
- Example: text, images, videos, observations
- Used for understanding behavior or patterns
- Many projects use both types to get better insights.
Sources of Data
Data can come from:
- Human observations
- Experiments
- Field notes or journals
- Lab results
- Direct observation of people, animals, or plants
Questions to Ask Before Selecting Data
- What is the research question?
- What is the scope of the study?
- What type of data is needed (quantitative, qualitative, or both)?
- What do previous studies suggest?
- Is the required data available?
Feature Selection in Data Mining
Feature selection means choosing only the most important variables
(features) from a dataset and removing unnecessary ones.
It helps improve model performance and reduces complexity.
Simple Example
- A doctor uses only important medical test results (features) to decide whether surgery is needed.
Types of Feature Selection
1. Supervised Learning
- Uses labeled data
- Goal: improve prediction accuracy
2. Unsupervised Learning
- No labeled data
- Goal: find better grouping (clustering)
Why Feature Selection is Important
Feature selection helps to:
- Reduce extra or useless data
- Improve model accuracy
- Make models easier to understand
- Save time, memory, and computing power
Problems Without Feature Selection
- Too much data → slows down processing
- Noisy or duplicate data → reduces accuracy
- High-dimensional data → needs more training data
How Feature Selection Works
- Each feature is given a score
- Important features are selected
- Less useful features are removed
This can be done:
- Manually by analysts
- Automatically by algorithms
Feature Selection in SQL Server
In SQL Server Data Mining:
- Feature selection happens before training the model
- The system calculates a score for each feature
- Only high-scoring features are used
Factors affecting selection:
- Algorithm used
- Type of data
- Model settings
Feature Selection Scoring Methods
1. Interestingness Score
Measures how useful a feature is
Based on entropy (information content)
Less randomness = more useful feature
2. Shannon’s Entropy
Measures uncertainty in data
Formula:
H(X) = -Σ P(x) log(P(x))
3. Bayesian (K2 Algorithm)
Uses probability and relationships between variables
Works well with structured data
4. Bayesian Dirichlet (BDE)
Uses probability distributions
Assumes equal importance of prior data
Feature Selection Parameters
These control how many features are selected:
MAXIMUM_INPUT_ATTRIBUTES
→ Limits number of input features
MAXIMUM_OUTPUT_ATTRIBUTES
→ Limits number of output features
MAXIMUM_STATES
→ Limits number of values in a feature
If set to 0, feature selection is turned OFF.
Conclusion
Data Selection ensures you use the right data for your problem
Feature Selection ensures you use only the most important data
Together, they improve:
- Accuracy
- Efficiency
- Model performance
