Importance of Characteristic Features and Their Form for Data Exploration.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Input feature characteristics critically influence the choice and performance of knowledge discovery tools and methods.
• Variable types, domains, and their relevance impact data exploration effectiveness and may necessitate preprocessing.
• Feature selection and reduction techniques, like ranking, are crucial for estimating attribute importance.

Purpose of the Study:

• To investigate the impact of feature relevance and data discretization on knowledge discovery performance.
• To propose and evaluate a procedure for gradual discretization controlled by attribute ranking.
• To assess the effectiveness of this approach in the domain of stylometry and authorship attribution.

Main Methods:

• Employed supervised and unsupervised discretization methods.
• Utilized attribute ranking for controlled, gradual discretization.
• Applied the methods to datasets from the stylometric domain for binary authorship attribution.
• Conducted extensive tests with selected classifiers.

Main Results:

• Data form conditioning based on relevance through gradual discretization was implemented.
• Partially discretized datasets showed enhanced prediction accuracy in many cases.
• The proposed discretization procedure, guided by attribute ranking, proved effective.

Conclusions:

• Feature relevance and appropriate data transformation, such as guided discretization, are key to improving machine learning model performance.
• The proposed method offers a viable approach for enhancing authorship attribution accuracy.
• Understanding and manipulating feature characteristics is vital for successful knowledge discovery.