AI and ML for Geodata Analysis: Session 2 Notes

Introduction

• Today's topic: Machine Learning Algorithms
• Presenter: Dr. Punam Sayari
• Reminder for participants to watch the session via YouTube and complete quizzes for certification.

Importance of Data

• Quote from Professor GTH James: "For the 21st century, data is the sword; handling it properly makes one a samurai."
• Need for advanced algorithms to handle large data sets.

Definitions

Machine Learning (ML)

• Subset of Artificial Intelligence (AI)
• Automates analytical models using data.
• ML learns relationships between input/output data sets and predicts outcomes.

Deep Learning

• Subset of ML using deep neural networks.
• Utilizes complex layers for feature extraction and prediction.

Key Differences Between ML and Deep Learning

1. Problem-Solving Approach
   • ML: Features are manually extracted.
   • Deep Learning: Minimal human intervention; features are learned automatically.
2. Training Methods
   • ML: Supervised, unsupervised, reinforcement learning, etc.
   • Deep Learning: Uses specialized architectures (autoencoders, CNNs, RNNs).
3. Algorithm Complexity
   • ML: Varied complexity.
   • Deep Learning: Complex architectures of interconnected neurons.
4. Data Requirements
   • ML: Can work on smaller datasets.
   • Deep Learning: Requires large datasets and significant computational power.

Types of Machine Learning Algorithms

1. Supervised Learning

• Uses labeled data for training.
• Examples: Classification and regression tasks.
• Classification: Assigning labels to input data (e.g., land cover classification).
• Regression: Predicting continuous output values.

2. Unsupervised Learning

• No labeled data; finds patterns on its own.
• Clustering: Groups similar data points together.
• Examples: K-means clustering, hierarchical clustering.

3. Semi-supervised Learning

• Combines a small amount of labeled data with a large amount of unlabeled data.

4. Reinforcement Learning

• Software learns to make decisions through trial and error.
• Actions rewarded or penalized to optimize outcomes.

Common Machine Learning Algorithms

Supervised Learning Algorithms

1. Paralle Classification Algorithm: Fast and simple classification method based on means and standard deviations.
2. Minimum Distance to Means Classification: Assigns pixels based on the shortest distance to class means.
3. Mahalanobis Decision Rule: Considers covariance, effective for overlapping classes.
4. Maximum Likelihood Classification: Based on probability distributions, most accurate but computationally intensive.

Decision Tree Classifier

• Makes binary decisions to classify data.
• Easy to understand and interpret.

Random Forest Classifier

• Ensemble method using multiple decision trees to improve accuracy and reduce overfitting.

Support Vector Machines (SVM)

• Finds optimal hyperplanes for classification tasks; effective in high-dimensional spaces.

Artificial Immune Networks & Logistic Regression

• Used for complex computational problems and basic classification tasks.

Unsupervised Learning Algorithms

• K-means Clustering: Assigns data points to clusters based on distance from centroids.
• Advantages: Automatically groups data based on similarities.
• Disadvantages: Less accuracy due to unlabeled data.

Conclusion

• Importance of selecting the right algorithm based on the problem, data, and desired outcomes.
• Session will continue with a Q&A session following a short break.