lecture 7

Data Mining for eHealth - Comprehensive Notes

Data Mining Overview

• Data Mining: "The process of discovering patterns in data. The process must be automatic or (more usually) semi-automatic. The patterns discovered must be meaningful." (Witten and Frank, 2005)

• Data mining focuses on extracting meaningful information from medical data

• Various types of medical data can be collected in healthcare environments

• There's a growing abundance of data through connected devices (IoT)

• A knowledge gap exists between data availability and systems for data analytics

Knowledge Discovery in Databases (KDD) Process

• KDD (Knowledge Discovery in Databases): A structured process that involves data collection, preprocessing, data mining techniques application, and evaluation/interpretation to extract knowledge from data.

The process includes:

1. Data Collection: Gathering relevant datasets

2. Preprocessing:

   • Dealing with missing values

   • Creating additional attributes

   • Performing attribute selection

3. Data Mining Techniques: Applying algorithms to extract patterns

4. Evaluation and Interpretation: Analyzing results to gain knowledge

Data Structure in Healthcare

• Structured Data:

  • Rows = Records/Examples/Instances (e.g., patient data)

  • Columns = Features/Attributes (e.g., age, symptoms)

  • Example attributes from breast cancer study: tumor location, size, age at diagnosis, irradiated status, recurrence

• Attribute Selection:

  • In theory: more attributes should result in more accurate patterns

  • In practice: irrelevant data may "confuse" algorithms

  • Selection methods:

    • Manual selection: using domain knowledge from experts

    • Automatic selection: reduces dimensionality by deleting unsuitable attributes based on heuristics

• Attribute Construction:

  • Creating new attributes to make regularities more apparent

  • Uses simple relational rules to reduce complexity in datasets

  • Can achieve dimensionality reduction

Data Mining Tasks

Descriptive Data Mining

• Clustering: The process of identifying groups where data points within a cluster are similar while those across clusters are dissimilar.

  • Finds finite sets of categories (clusters) to describe data

  • Groups examples so similarity within clusters is maximized and similarity between clusters is minimized

  • Applications:

    • Market basket analysis (items bought together)

    • Healthcare: identifying patients with similar treatment needs

  • Example: Grouping patients based on LDL, HDL, and BMI measurements

Predictive Data Mining

• Regression: Finding a model (function) that maps a given input (attributes values) to a numeric prediction.

  • Deals with time components, crucial for forecasting data

  • Applications:

    • Forecasting economy growth based on market indicators

    • Predicting patient recovery trajectory after injury or surgery

• Classification: Finding a model that is able to predict the value of the class attribute of an example based on the values of a set of attributes.

  • Each record belongs to a predefined class

  • Each example consists of predictor attributes and a class attribute

  • Requires training sets for model building and test sets for evaluation

  • Goal: discover relationships that allow prediction of class values

  • Process flow:

    1. Use training set (known-class examples) to build a model

    2. Apply model to test set (unknown-class examples) to predict classes

Classification Algorithms in Healthcare

• Artificial Neural Networks (ANNs):

  • Inspired by complex neural connections in biological systems

  • Built from densely interconnected simple units

  • Each unit takes real-valued inputs and produces a single real-valued output

  • Well-suited for noisy, complex sensor data often found in healthcare contexts

• Support Vector Machines (SVMs):

  • Binary linear classifier

  • Goal: find the maximum margin hyperplane (plane with greatest separation between classes)

  • Identifies support vectors to define decision boundaries

• Decision Trees:

  • Comprehensible graphical representation of a classification model

  • Internal nodes correspond to attribute tests (decision nodes)

  • Leaf nodes correspond to predicted class values

  • Classification process:

    • Tree is traversed top-down from root to leaf

    • Branches selected according to attribute test outcomes

    • Class value at leaf node is the predicted class

• Rule Induction (was mentioned in your notes but not detailed in the provided material)

Importance of Interpretability and Trustworthiness

• Crucial for adoption of predictive systems in healthcare

• Need for clear reasoning behind model predictions to build trust with healthcare professionals

• Decision trees offer better interpretability compared to "black box" models like neural networks

Additional Information

• The "Bioinformatics Knowledge Gap" chart shows the increasing disparity between available data and our ability to extract knowledge from it

• The lecture mentions specific evaluation methods for classification model performance will be discussed in a future class

• Attribute maintenance is important (e.g., using date of birth instead of age for better data management)