Novel Algorithms and Applications for Data Mining and Machine Learning
Sudipta Pathak · OpenCommons at University of Connecticut (University of Connecticut) · 2020
In this dissertation we propose novel approaches for data mining and machine learn- ing for some of the fundamental and advanced problems in the areas of data analysis and bioinformatics. A number of problems such as Drug-Target Interaction (DTI) pre- diction, Singular Value Decomposition (SVD), Time series Analysis (TSA) and motif search has been studied in this thesis. We have developed algorithms that outperformed the state of art in all of the above mentioned areas. We proposed algorithms for DTI prediction that outperformed all prior algorithms over benchmark datasets under mul- tiple scenarios. Our proposed approaches for Jacobi based SVD, both sequential and parallel, improves the computation time and work compared to the state of art. We also proposed ensemble based TSA models that improves classification accuracy sta- tistically significantly compared to all prior algorithm on benchmark dataset. We also introduced novel motif search algorithms for DNA and protein motifs.