|
Our course "Data Mining in Finance - Predictive Analytics and Financial
Forecasting" provides an introduction into data mining methods and
demonstrates how data mining can be employed in various tasks in the
financial sector by banks, investment funds, hedge funds, insurance
companiess, and other financial institutions and companies in the finance
sector as well as sophisticated private investors and traders.
Financial data often comes in the form of time serieses, e.g. stock market
prices, commodity prices, utility prices, or currency exchange rates
observed over time.
The analysis of financial time series is often based on indicators
derived from such uni- or multi-variate time series and on linear or
non-linear combinations of such indicators.
Data mining techniques can be used to analyse financial time series data,
to find patterns, to detect anomalies and outliers, to recognize situations
of chance and risk, to detect temporal changes in the correlation patterns
and structures, to predict future demand, prices, and rates, to determine
the most successful indicators, to optimally combine such indicators to
achieve strong predictive power.
Hence data mining can support analysts, investors, and traders in their
decisions when trading stocks, options, commodities, utilities, or
currencies.
This course describes data mining techniques for financial time series
data including data pre-processing methods, time window based approaches
and the automated optimization of the time window sizes, indicator or
feature weighting and selection methods, and statistical and machine
learning techniques to automatically find weighted linear and non-linear
combination of an optimized set of indicators or features when
automatically learning a prediction or forecasting model (trade decision
(support) system).
The topics include necessary preprocessing steps for numerical data
transformations, an introduction into linear and non-linear statistical
regression methods, neural networks, and support vector machines (SVM),
and a discussion of validation methods in order to measure the goodness
of the predictions, i.e. how to thoroughly evaluate such models on
historic data (backtesting).
These methods are especially useful for numerical predictions from
series data as they often occur in financial markets.
In addition to time series data, transactional data is also very frequent
in the financial sector.
Transactional and accounting data can be analyzed and leads to models for
detecting patterns and irregularities, improper trading or accounting
practices, dubios transactions, potential fraud, money laundering, and other
undesired activities as well as for transaction monitoring, credit scoring,´
credit default prediction, risk assessment and minimization, finding risk
factors, and portfolio management.
This course also describes the practical steps necessary to create such
models with the open-source data mining software RapidMiner. The large
amount of practical examples and exercises enables the participants to
design appropriate data mining processes and apply the gained knowledge to
their data mining problems and to solve them efficiently and successfully.
|