Classifying
Streaming Data
Week 8-9
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Guidelines and tips
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Data Mining Additional Discussion
Typology: Cheat Sheet
2021/2022
1 / 33
This page cannot be seen from the preview
Don't miss anything!
Related documents
Partial preview of the text
Download Data Mining Additional Discussion and more Cheat Sheet Technology in PDF only on Docsity!
Classifying
Streaming Data
Week 8-
What is Streaming Data?
- (^) Data arriving continuously and rapidly
- (^) Example of Streaming data are :
- (^) Internet of things
- (^) Social Media
Key Challenges :
- (^) High velocity
- (^) Memory constraints
- (^) Concept drift
- (^) Limited access
Challenges in Classifying Streaming Data
- (^) High Velocity : Need for real-time or near-real-time processing.
- (^) Memory Constraints : Cannot store entire datasets.
- (^) Concept Drift : Data distributions change over time.
- (^) Imbalanced Data : Uneven class distributions in streams.
Popular Algorithms
- (^) Hoeffding Tree (H-Tree) : Incremental decision tree with low memory requirements.
- (^) Naïve Bayes :Lightweight probabilistic classifier.
- (^) Stochastic Gradient Descent (SGD) :Efficient for online updates.
- (^) Online Random Forest :Adaptive ensemble of decision trees.
Handling Concept Drift
- (^) Concept Drift
- (^) Defined : Changes in data distribution affecting model performance.
- (^) Strategies :
- (^) Sliding Window: Focus on recent data.
- (^) Weighted Learning: Prioritize newer data.
- (^) Drift Detectors: Detect and adapt to drift (e.g., DDM, EDDM).
Applications
- (^) Fraud Detection :
- (^) Analyzing live transactions for fraud.
- (^) IoT Monitoring :
- (^) Real-time sensor event classification.
- (^) Social Media Analytics :
- (^) Sentiment or trend analysis.
- (^) Traffic Monitoring :
- (^) Predicting congestion or accidents.
Tools for Implementation
- (^) MOA : Massive Online Analysis framework.
- (^) Apache Flink : Distributed processing for streams.
- (^) scikit-multiflow : Python library for streaming data.
- (^) River : Lightweight library for online machine learning.
Hoeffding Tree (H-Tree)
Hoeffding Tree (H-Tree)
- (^) also known as an H-Tree or Very Fast Decision Tree (VFDT)
- (^) a type of decision tree algorithm designed specifically for streaming data.
- (^) It is a machine learning model that can efficiently handle large datasets by incrementally learning from a stream of data, rather than requiring all data to be available at once.
Characteristics of Hoeffding Tree
- (^) Hoeffding Bound :
- (^) The algorithm uses the Hoeffding bound (a statistical measure) to decide when to split a node in the tree.
- (^) The bound provides a guarantee on the minimum number of samples required to make confident decisions about the best attribute to split on.
Characteristics of Hoeffding Tree
- (^) Efficiency :
- (^) Suitable for streaming and large datasets where storing the entire dataset in memory is impractical.
- (^) Works efficiently even when the data arrives at high velocity.
Algorithm Overview (H-Tree)
- (^) Node Splitting :
- (^) Nodes are split based on the best attribute determined using a splitting criterion (e.g., information gain or Gini index).
- (^) Instead of calculating the criterion over the entire dataset, the Hoeffding bound ensures that the decision is statistically sound with a subset of the data.
Algorithm Overview (H-Tree)
- (^) Attributes and Statistics :
- (^) The algorithm maintains sufficient statistics (e.g., counts, sums) at each node to compute splitting criteria.
- (^) Only attributes that meet the Hoeffding bound confidence level are used to split.