Data Mining and Neural Networks

Module code: MA7022

This module will provide a comprehensive grounding in the structure of the data mining process and explain the basic notations and operations. During this module you will familiarise yourself with types of data mining problems and select the necessary approach to its solutions, from evaluation and cleaning of the dataset to selection of the algorithms for data analysis. This module will also cover the basic methods and algorithms for data analysis, and how to construct basic neural networks for data analysis.

Topics covered

  • Classification (knn and Decision tree algorithms)
  • Entropy and conditional entropy
  • Information gain, clustering (k-means, hierarchical clustering and density based algorithms)
  • Multivariate regression (linear regression and the kernel trick)
  • Probability distribution estimation (Bayes rule and Bayes networks – structure and algorithm of construction)
  • Preprocessing
  • Data cleaning
  • Dimension reduction (principal component analysis)
  • Time series, stationary time series (in strong and in weak sense), white noise, random walk, moving average processes
  • Autoregressive processes
  • Integrated and ARIMA processes
  • Smoothing -mean filter and median filter
  • Trend analysis and segmentation
  • Formal neurons
  • Perceptron learning rule and Widrow-Hoff learning rule
  • Main types of neural networks (Hopfield, Kohonen and back- propagation of errors) and problems they can solve
  • Cascade-correlation algorithms

Learning

  • 33 one-hour lectures
  • 11 one-hour seminars
  • 11 one-hour workshops
  • 95 hours of guided independent study

Assessment

  • Exam, two hours (50%)
  • Problem sheets (20%)
  • Computational tasks (30%)