Data mining finds patterns in data — but raw data is messy. Learn the data-mining functionalities and the preprocessing that makes data ready: cleaning, integration, transformation, reduction and discretization, with a live normalization playground.
Data mining is the process of discovering interesting, previously unknown and useful patterns from large amounts of data — it's the core analysis step of the broader KDD (Knowledge Discovery in Databases) process.
Functionalities describe what can be mined: characterization & discrimination, association & correlation, classification & regression (prediction), cluster analysis, and outlier analysis. These split into descriptive (summarise the data) and predictive (forecast unknown values) tasks.
Mining is one step in a pipeline. Most of the effort lives in preparing the data.
Different features come in different ranges; many algorithms need them on a common scale. Pick a method and watch the same seven values rescale — the bars and the numbers update live with the real formula.
Characterization, association, classification, clustering, outlier & evolution analysis.
Describe patterns in current data vs predict unknown/future values.
What to mine: data set, knowledge type, background knowledge, interestingness measures, presentation.
No-coupling, loose, semi-tight, tight coupling with a DB/DW system.
Methodology, user interaction, efficiency & scalability, diversity of data, applications & social impact.
Fill missing values, smooth noisy data (binning, regression), remove outliers, fix inconsistencies.
Combine multiple sources; resolve schema/entity conflicts and redundancy (correlation tests).
Normalization, aggregation, generalization, attribute construction.
Dimensionality reduction (PCA), numerosity reduction, data compression — smaller yet representative.
Turn continuous values into intervals; build concept hierarchies (city→state→country).