FREQUENT PATTERN

MINING FROM
DATA STREAM
 PRESENTED BY:
Rajeev M. – AP24122040003
Tejashri A. – AP24122040009
Ridhamkumar T. – AP24122040014
Suleiman Ismail – AP24122040020
P. Sandeep –AP24122060005
G.Sri Harsha Vardhan-AP24122060007
K.V. Krishna Sampath- AP24122060017
 WHAT IS DATA
STREAM?
● A continuous, high-speed, and time-varying flow of data
generated in real-time.
● Unlike traditional datasets, data streams are unbounded,
evolving, and cannot be stored entirely.
● Challenges: Memory constraints, concept drift, and scalability.
● Examples:
Sensor networks (e.g., IoT devices)
Online transactions (e.g., stock market, e-commerce)
Social media feeds (e.g., Twitter, Facebook)
KEY CHARACTERISTICS OF DATA
STREAM
● Unbounded
The data continuously flows and never ends.
● High-speed
Data arrives rapidly, requiring real-time processing.
● Concept Drift
The patterns and distributions in data can change over
time.
● One-pass Processing
Data is processed only once due to memory constraints.
CHALLENGES IN MINING DATA
STREAMS
● Memory Constraints
Traditional algorithms assume data fits in memory, but
streaming data is too large.

● Scalability
Must handle large volumes of fast-arriving data efficiently.

● No Reprocessing
Once data passes, it cannot be re-evaluated.
WHAT IS FREQUENT PATTERN
MINING?
● Frequent Pattern Mining(FPM) is a data mining technique
used to discover recurring patterns, associations, or
correlations in large datasets.

● Initially used in market basket analysis, where it identifies

items frequently bought together.

● Applied in fraud detection, network security,

bioinformatics, and recommendation systems.
KEY CONCEPTS IN FPM
● Frequent Itemsets
Sets of items that appear frequently together in
transactions.
● Support
The proportion of transaction containing an itemset.
● Confidence
The likelihood that one item appears given another (used in
association rules).
● Lift
Measures how much more likely two items occur together
compared to if they were independent.
CHALLENGES IN FPM
● High Computational Cost
Exponential number of possible itemsets.

● Handling Dynamic Data

Frequent patterns change over time in data streams.

● Memory Limitations
Processing large-scale datasets requires efficient storage
structures.
APPLICATIONS OF FPM
● E-commerce
Personalized recommendations based on past purchases.

● Cybersecurity
Detecting abnormal behavior in network traffic.

● Healthcare
Identifying disease correlations from patient records.
POPULAR ALGORITHMS FOR FPM

1. Apriori Algorithm

2. FP-Growth (Frequent Pattern Growth)

Algorithm

3. Lossy Counting Algorithm

1. APRIORI ALGORITHM
INTRODUCTION

● A data mining algorithm used to find frequent itemsets and

generate association rules.

● Helps identify relationships between items in large datasets,

widely used in market basket analysis.

● Example: If customers buy bread, they often buy butter →

Helps in product placement & marketing.
1. APRIORI ALGORITHM
HOW IT WORKS?
● Identifying Frequent Itemsets
Scans the dataset, finds single items & applies a minimum
support threshold.
● Creating Candidate Item Groups
Combines frequent items into larger itemsets iteratively.
● Pruning Infrequent Item Groups
Uses the Apriori Property (if an itemset is infrequent, all
its supersets are infrequent).
● Generating Association Rules
Extracts relationships using support, confidence, and
lift.
1. APRIORI ALGORITHM
KEY MATRICS
● Support
Frequency of an itemset in all transactions ("Bread
appears in 20% of purchases").
● Confidence
Probability of buying Y when X is bought ("If bread is
bought, butter follows 75% of the time").
● Lift
Measures how strongly two items are associated ("Bread &
butter are bought together more than by chance").
1. APRIORI ALGORITHM
NUMERICAL EXAMPLE

Let's consider a small dataset with 4 transactions in a grocery

store:

Datasets (Transactions)
TRANSACTION ID ITEM PURCHASED

T1 Milk, Bread

T2 Milk, Butter

T3 Bread, Butter

T4 Milk, Bread
1. APRIORI ALGORITHM
Step 1 : Identifying Frequent Itemsets (Support Calculation)
● Let’s set minimum support = 50% (≥2 transactions out of 4).
● 1-itemsets (Support Count)
● Milk = 3/4 = 75% ✅
● Bread = 3/4 = 75% ✅
● Butter = 2/4 = 50% ✅

● All 1-itemsets meet the minimum support, so we move to 2-itemsets.

● 2-itemsets (Support Count)
● (Milk, Bread) = 2/4 = 50% ✅
● (Milk, Butter) = 1/4 = 25% ❌ (Below 50%, discard)
● (Bread, Butter) = 1/4 = 25% ❌ (Below 50%, discard)
● Final frequent itemsets:
✅ {Milk}, {Bread}, {Butter}, {Milk, Bread}
1. APRIORI ALGORITHM
Step 2 : Generating Association Rules (Confidence Calculation)
● Using minimum confidence = 60%, we generate rules:

● Rule 1: {Milk} → {Bread}

● Confidence = (Milk, Bread) / Milk = (2/3) = 66.7% ✅

● Rule 2: {Bread} → {Milk}

● Confidence = (Milk, Bread) / Bread = (2/3) = 66.7% ✅
● Rule 3: {Bread} → {Butter}
● Confidence = (Bread, Butter) / Bread = (1/3) = 33.3% ❌
● Rule 4: {Milk} → {Butter}
● Confidence = (Milk, Butter) / Milk = (1/3) = 33.3% ❌
1. APRIORI ALGORITHM
Step 3 : Lift Calculation (For {Bread} → {Milk})
● Lift = Confidence / Support of Bread
● Lift = 66.7% / 75% = 0.88

Final Association Rules:

✅ {Milk} → {Bread} (66.7% confidence)

✅ {Bread} → {Milk} (66.7% confidence)
1. APRIORI ALGORITHM
Advantages:
● Simple and easy to implement.
● Works well for small datasets.

Limitations:
● Expensive in computation (requires multiple database scans).Works well
for small datasets.
● Generates many candidate itemsets, increasing complexity.

Applications:
● Market Basket Analysis (e.g., finding products frequently bought
together).
● Web usage mining (e.g., analyzing browsing behavior).
2. EQUIVALENCE CLASS
CLUSTERING AND BOTTOM-UP
LATTICE TRAVERSAL (ECLAT)
INTRODUCTION

● It is a more efficient and scalable version of the Apriori

algorithm.

● It uses vertical data format (item → list of transaction IDs) to

identify frequent itemsets.

● ECLAT is efficient and fast, especially for dense datasets.

2. ECLAT ALGORITHM
HOW IT WORKS?
● Convert transactions into a vertical database — for each
item, store a list of transactions (TID-list) where the item
appears.
● Find frequent 1-itemsets using support threshold.
● Generate larger frequent itemsets by intersecting TID-lists of
smaller itemsets.
● If intersection count (support) ≥ minimum support threshold,
keep the itemset; otherwise, discard.
2. ECLAT ALGORITHM
KEY MATRICS
● Support
Number of transactions containing the itemset.

● TID-list
List of transaction IDs where an item occurs.

● Minimum Support
Threshold to decide if an itemset is frequent.
2. ECLAT ALGORITHM
NUMERICAL EXAMPLE

Let's consider a small dataset with 4 transactions in a grocery

store:

Datasets
TRANSACTION ID (Transactions)
ITEM PURCHASED

T1 Milk, Bread

T2 Milk, Butter

T3 Bread, Butter

T4 Milk, Bread
2. ECLAT ALGORITHM
Step 1 : Create Vertical Data Format (TID-lists)

ITEM TID-list

Milk {T1, T2, T4}

Bread {T1, T3, T4}

Butter {T2, T3}

Step 2 : Find Frequent 1-itemsets

● Assuming Minimum Support = 2
● All items (Milk, Bread, Butter) are frequent as they each appear ≥ 2 times.
2. ECLAT ALGORITHM
Step 3 : Find Frequent 2-itemsets via TID-list Intersections

ITEMSET TID-list Intersection SUPPORT

{Milk, Bread} {T1, T4} 2

{Milk, Butter} {T2} 1(discard)

{Bread, Butter} {T3} 1(discard)

Step 4 : Find Frequent 3-itemsets

● {Milk, Bread, Butter}: Intersection of {T1, T4} ∩ {T2, T3} = ∅ (support = 0,
discard).
2. ECLAT ALGORITHM
Step 5 : Final Frequent Itemsets

ITEMSET SUPPORT

{Milk} 3

{Bread} 3

{Butter} 2

{Milk, Bread} 2
2. ECLAT ALGORITHM
Advantages:
● Fast for dense datasets.
● Uses vertical format, making intersections efficient.
● Requires fewer scans of the database.
● Can handle long frequent patterns effectively.

Limitations:
● Not suitable for sparse datasets (many empty intersections).
● Memory intensive when TID-lists are very large.
● Performs poorly when itemsets are very long in high-dimensional data.

Applications:
● Market Basket Analysis — to find products frequently bought together.
● Web usage mining — to discover frequently visited pages.
● Bioinformatics — to detect frequent patterns in gene sequences.
● Recommendation Systems — to suggest frequently grouped items.
3. FREQUENT PATTERN(FP)
GROWTH ALGORITHM
INTRODUCTION

● Frequent Pattern Growth (FP-Growth) is an efficient algorithm

for frequent pattern mining.

● It overcomes the drawbacks of the Apriori Algorithm by

avoiding multiple database scans and eliminating the need to
generate candidate sets.

● FP-Growth avoids the inefficiencies of the Apriori Algorithm

by using a tree-based approach to mine frequent patterns
efficiently.
3. FREQUENT PATTERN(FP)
GROWTH ALGORITHM
HOW IT WORKS?
● Build the FP-Tree
Scan the dataset once to count item frequencies.
Construct a compact FP-Tree by grouping transactions
with common items.
● Extract Frequent Patterns
Traverse the FP-Tree bottom-up.
Generate conditional FP-Trees to find frequent itemsets.
● Generating Association Rules
Identify relationships between frequent itemsets.
Use support and confidence to validate the rules.
3. FREQUENT PATTERN(FP)
GROWTH ALGORITHM
KEY MATRICS
● Support
Frequency of an itemset in all transactions ("Bread
appears in 20% of purchases").
● Confidence
Probability of buying Y when X is bought ("If bread is
bought, butter follows 75% of the time").
● Lift
Measures how strongly two items are associated ("Bread &
butter are bought together more than by chance").
3. FREQUENT PATTERN(FP)
GROWTH ALGORITHM
NUMERICAL EXAMPLE

Let's consider a small dataset with 4 transactions in a grocery

store:

Datasets (Transactions)
TRANSACTION ID ITEM PURCHASED

T1 Milk, Bread

T2 Milk, Butter

T3 Bread, Butter

T4 Milk, Bread
3. FREQUENT PATTERN(FP)
GROWTH ALGORITHM
Step 1 : Count Frequency of Items
● Let’s count how many times each item appears:

ITEM FREQUENCY(SUPPORT COUNT)

Milk 3

Bread 3

Butter 2

● Assume Minimum Support = 2, so all items are frequent.

3. FREQUENT PATTERN(FP)
GROWTH ALGORITHM
Step 2 : Sort Items in Each Transaction (by frequency)

TRANSACTION ID ITEM PURCHASED

T1 Bread, Milk

T2 Milk, Butter

T3 Bread, Butter

T4 Bread, Milk
3. FREQUENT PATTERN(FP)
GROWTH ALGORITHM
Step 3 : Build FP-Tree
Roo
t

Milk Bread
(1) (3)

Butter Milk Butter

(1) (2) (1)
3. FREQUENT PATTERN(FP)
GROWTH ALGORITHM
Step 4 : Final Frequent Itemsets

FREQUENT ITEMSET SUPPORT COUNT

{Bread} 3

{Milk} 3

{Butter} 2

{Bread, Milk} 2
3. FREQUENT PATTERN(FP)
GROWTH ALGORITHM
Step 5 : Generate Association Rules
From frequent itemsets:
1. Bread → Milk
● Support = 2/4 = 50%
● Confidence = 2/3 ≈ 66.7%
2. Milk → Bread
● Support = 2/4 = 50%
● Confidence = 2/3 ≈ 66.7%

Final Output:
Frequent Itemsets:
● {Bread}, {Milk}, {Butter}, {Bread, Milk}
Association Rules:
● Bread → Milk (Confidence: 66.7%)
● Milk → Bread (Confidence: 66.7%)
3. FREQUENT PATTERN(FP)
GROWTH ALGORITHM
Advantages:
● Faster than Apriori (No need to generate candidate sets).
● Efficient for large datasets (Uses a compact FP-Tree).
● Reduces database scans (Processes data in a compressed way).

Limitations:
● Tree Construction Overhead (Consumes memory for large datasets).
● Not scalable for extremely large databases.

Applications:
● Market Basket Analysis → Recommending products based on purchase
history.
● Fraud Detection → Identifying unusual transaction patterns.
● Web Usage Mining → Analyzing user behavior for recommendations.
● Bioinformatics → Identifying common gene sequences.
REFERENCE
● Han, J., Pei, J., & Kamber, M. (2011). Data Mining: Concepts and Techniques (3rd
ed.). Morgan Kaufmann.

● Agrawal, R., & Srikant, R. (1994). Fast algorithms for mining association rules.
Proceedings of the 20th International Conference on Very Large Data Bases
(VLDB), 487–499.

● Han, J., Pei, J., & Yin, Y. (2000). Mining frequent patterns without candidate
generation. ACM SIGMOD Record, 29(2), 1–12.

● Zaki, M. J., Parthasarathy, S., Ogihara, M., & Li, W. (1997). New algorithms for
fast discovery of association rules. KDD, 283–296.

● Manku, G. S., & Motwani, R. (2002). Approximate frequency counts over data
streams. Proceedings of the 28th International Conference on Very Large Data
Bases (VLDB), 346–357.

● https://www.geeksforgeeks.org/
Thank
You!