<p>The self-organizing sequential search problem, known as the List Update Problem (LUP), represents a fundamental challenge in algorithmic design under the online framework where complete request sequence is unknown. This paper presents a comprehensive theoretical as well experimental investigation spanning behavioural insights of Move-to-Front-or-Logarithmic Position (MFLP) on different class of request sequence, novel adaptive framework under online framework, and machine learning-driven strategy to predict optimal reorganization strategy. The three phase investigation, first focuses on formulating theoretical cost bound of MFLP on different characterization of request sequences. The results shows that for request sequences exhibiting lower degree of locality of references, MFLP performs better than well known MTF, which is further evaluated through experimental results on Calgary and Canterbury corpus datasets. In the second phase of investigation, two new adaptive algorithms are proposed known as Context-Based Adaptive Reorganization (CBAR) strategy, and its extension, the Hybrid Context-Based Adaptive Reorganization (HCBAR). The experimental evaluation of these algorithms depict that CBAR outperforms all other algorithms with the winning percentage of 66.7 when they are operated on 18 different types of synthetic data sequences. In the third phase, this attempts for the first time to design machine learning based approach to predict optimal rearrangement strategy in the field of self organizing sequential search. A four-stage hierarchical ensemble architecture is developed, achieving state-of-the-art classification accuracy of 88.25% in predicting the most effective reorganization strategy for arbitrary request sequences.</p>