1st question

Frequent Itemset Generation Using Apriori Algorithm

What It Does: The Apriori Algorithm nds items that are often bought together in transactions.

Steps

1. Support: Check how often an item or group of items appears in transactions.

Support
=
Number of Transactions with the Item(s)
Total Transactions

\text{Support} = \frac{\text{Number of Transactions with the Item(s)}}{\text{Total Transactions}}

2. Threshold: Only keep items that appear frequently enough (e.g., at least 40%).

Example

Dataset:

Transactio
Items Bought
n
T1 Bread, Milk, Beer
Bread, Diaper,
T2
Milk
T3 Milk, Diaper, Beer
T4 Bread, Milk
T5 Bread, Diaper

1. Find Single Frequent Items:

◦ Bread: 4/5 = 0.8

◦ Milk: 4/5 = 0.8
◦ Beer: 2/5 = 0.4
◦ Diaper: 3/5 = 0.6
Keep: Bread, Milk, Beer, Diaper.
2. Find Frequent Pairs:

◦ Bread + Milk: 3/5 = 0.6 (Keep)

◦ Bread + Diaper: 2/5 = 0.4 (Keep)
◦ Milk + Diaper: 2/5 = 0.4 (Keep)
3. Find Frequent Triples:
fi
 ◦ Bread + Milk + Diaper: 1/5 = 0.2 (Too low, discard).

Result:

• Frequent Single Items: Bread, Milk, Beer, Diaper.

• Frequent Pairs: {Bread, Milk}, {Bread, Diaper}, {Milk, Diaper}.

Conclusion: The Apriori Algorithm shows patterns like:

"If someone buys Bread, they are likely to buy Milk."

2nd question

2nd Question: FP-Growth Algorithm

What It Does: FP-Growth nds frequent itemsets without generating candidates like Apriori. It uses a
special tree (FP-Tree) to store transactions compactly.

Steps

1. Build the FP-Tree:

◦ Count how often items appear.

◦ Keep items meeting the support threshold (e.g., 40% or 2 transactions).
◦ Sort items by frequency and add transactions to the tree.
2. Mine the FP-Tree:

◦ Start from the bottom of the tree and nd combinations of frequent items.

Example:

Transactio
Items Bought
n
T1 Bread, Milk, Beer
Bread, Milk,
T2
Diaper
T3 Milk, Diaper, Beer
T4 Bread, Milk
T5 Bread, Diaper

Frequent Items: Bread (4), Milk (4), Diaper (3), Beer (2).

FP-Tree:

NULL
fi
fi
 |
Bread (4)
/ \
Milk (3) Diaper (1)
| |
Beer (1) Milk (1)
Frequent Itemsets:

• {Bread}, {Milk}, {Bread, Milk}, {Bread, Diaper}, {Milk, Diaper}.

Result: FP-Growth quickly nds frequent itemsets like:

• Bread and Milk often appear together.

3rd Question: What is Classi cation?

What It Does: Classi cation assigns categories (labels) to data based on past examples.

Examples:

• Spam Detection: Is an email "spam" or "not spam"?

• Medical Diagnosis: Is a tumor "benign" or "malignant"?

Issues in Classi cation

1. Imbalanced Data: If one class (e.g., "spam") is rare, the model may focus on the common class.

◦ Solution: Balance the dataset.

2. Over tting: The model learns noise in training data.

◦ Solution: Simplify the model or use more data.

3. Under tting: The model is too simple and misses patterns.

◦ Solution: Use a better algorithm or features.

4. Noisy Data: Errors or irrelevant information can confuse the model.

◦ Solution: Clean the data.

Conclusion: Classi cation helps predict labels, but good data and careful modeling are essential.

4th question

4th Question: Classi cation by Decision Tree Induction

What It Does: A Decision Tree splits data into branches based on rules to classify it into categories.
fi
fi
fi
fi
fi
fi
fi
fi
Steps

1. Feature Selection: Choose the best feature to split the data (e.g., "Weather").

◦ Use methods like Information Gain or Gini Index to decide.

2. Build the Tree:

◦ Split data into branches based on the chosen feature.

◦ Repeat until:
▪ All data in a branch belongs to the same class, or
▪ A stopping condition is met (e.g., max depth).
3. Prune the Tree:

◦ Remove unnecessary branches to avoid over tting.

Example

Dataset:

Weathe Temperatur Play

r e Tennis
Sunny Hot No
Sunny Mild No
Overcast Mild Yes
Rainy Cool Yes
Rainy Mild Yes

Decision Tree:

Weather?
├── Sunny: No
├── Overcast: Yes
└── Rainy: Yes
Prediction:

• If Weather is Sunny, predict No.

Advantages

• Easy to understand and visualize.

• Works with numerical and categorical data.
Disadvantages

• Can over t if the tree is too deep.

• Sensitive to noisy data.

Conclusion: Decision Trees classify data by splitting it based on features. They're simple but need pruning
to work well.
fi
fi
 Support Vector Machine (SVM)

• What It Does: SVM separates data points into classes using a line (or hyperplane in higher
dimensions).
• Key Idea: It nds the line that keeps the maximum distance from the closest points of each class
(called support vectors).
• When To Use: For data with clear separation between classes.
• Advantages:
◦ Works well with high-dimensional data.
◦ Effective when there’s a clear margin between classes.
• Disadvantages:
◦ Can be slow with large datasets.
◦ Needs careful selection of parameters (like kernels).

K-Nearest Neighbors (KNN)

• What It Does: KNN assigns a class to a point based on the majority class of its k

-nearest neighbors.
• Key Idea: It uses the distance to nearby points to classify new data.
• When To Use: Simple problems with smaller datasets.
• Advantages:
◦ Easy to understand and implement.
◦ No need for a training phase (lazy learning).
• Disadvantages:
◦ Can be slow for large datasets.
◦ Sensitive to irrelevant features and the value of k

Comparison of SVM and KNN

Feature SVM KNN

Model-based (builds a Instance-based (on-the-
Learning Type
rule) y)
Speed Fast prediction, slow train Slow prediction
Data
High-dimensional data Small datasets
Suitability

Both are useful, but SVM is better for complex and high-dimensional problems, while KNN is great for
simpler, intuitive tasks.
fl
fi