Development of a toolbox for data mining

Abstract

The subject of this thesis is the development of a Python application that implements basic algorithms for classification, regression, clustering, and association rules, all of which are considered fundamental in Data Mining. Existing implementations are integrated into a single app, accessible through a graphical environment with concise help of interactive nature. The field of Data Mining has roots in statistics, machine learning and pattern recognition theory. Ever since its emergence in the early 1990s, it has been at the epicenter of various methodologies that encompass it, notably KDD and CRISP-DM among many. The application supports loading a dataset, performing data cleaning and preprocessing, splitting the dataset for post-training evaluation, choosing one of the available data mining techniques, fitting the model, and, in the case of predictive models, the application of cross-validation as a final step (whether for model selection or for model evaluation). Models used in both classification and regression include Decision Trees, Support Vector Machines (SVM), and k-Nearest Neighbors (KNN). Available classifiers include Naïve Bayes and Logistic Regression. Additionally, there is Linear Regression for the regression category. Available clustering algorithms are K-Means, DBSCAN, and Agglomerative clustering. These implementations are based on the Scikit-learn library, with the app essentially acting as a wrapper or interface with its own additional features. For Association Rules, the Apriori and FP-Growth algorithms can be used for the generation of frequent itemsets. KivyMD was used for the front-end, which is based on Kivy, a GUI library available in Python. An appendix is dedicated to concepts and equations that are fundamental to the various subjects discussed in this thesis.