Classification and Clustering
| Feature | Classification | Clustering |
|---|---|---|
| Type of Learning | Supervised (data is assigned to predefined categories) | Unsupervised |
| Supervised/Unsupervised | Supervised | Unsupervised |
| Labeling Requirement | Requires labeled data | No predefined labels |
| Goal | Classifies new data into known categories | Discovers hidden structures in the data |
| Examples | Spam detection, disease diagnosis, sentiment analysis | Image clustering, anomaly detection |
| Algorithms | Decision Trees, SVM, Random Forest, Neural Networks | K-Means, DBSCAN, Hierarchical Clustering |
| Output | Discrete categories (e.g., spam or not spam) | Grouped clusters based on similarity |
| Use Case Focus | Predicting labels for new data | Finding patterns and natural groupings in data |
| Real-World Example | Email classification into spam/ham | Grouping customers by purchasing behavior |
KDD Issues
- Human Interaction
- Overfitting
- Outliers
- Interpretation because of the complexity of the model
- Visualization because of high dimensionality
- Large Data sets because of computational complexity
- MultiMedia data is an issue because format of the data can be in various forms.
- Example photos and videos
- Changing Data Streams is an issue because the model may not be able to adapt to the new data.
Assignment
Solve two numerical examples on on decision tree, regression, naivee baye’s, IDB
ID3
Entropy
If S is a collection of 14 examples with 9 yes and 5 no examples then Entropy(S)=-9/14 Log_2 (9/14) - (5/14)Log_2(5/14)=0.940
Information Gain
Gain is defined as the difference between how much information is needed to make a correct classification before the split versus how much information is needed after the split. Id3
References
Information
- date: 2025.03.12
- time: 13:14