Abstract :

Customer Retention Has Become A Major Concern For Service-oriented Businesses, As A Considerable Percentage Of Users Discontinue Services Each Year, Leading To Financial Losses And Increased Acquisition Costs. With Organizations Generating Vast Volumes Of Customer Data Such As Feedback, Complaints, Transaction Records, And Service Interactions, Extracting Meaningful Insights From Unstructured Text Has Become Essential. Traditional Churn Prediction Approaches, Including Manual Evaluation, Rule-based Systems, And Basic Analytical Methods, Often Fail To Capture Deeper Linguistic Patterns Such As Contextual Meaning, Sentiment Variation, And Fine-grained Textual Signals, Resulting In Limited Scalability And Inconsistent Outcomes. To Overcome These Challenges, This Study Proposes An Advanced Churn Prediction Framework That Integrates Character Architecture With No Tokenization In Neural Encoders (CANINE) Embeddings With Random Oblique Forest Trees (ROFT). For Comparative Analysis, Baseline Models Such As Ridge Classifier (RC) Nearest Centroid (NC), And A Hybrid Bernoulli RBM Combined With RC Are Also Evaluated. The Proposed Architecture Effectively Captures Both Semantic And Character-level Representations While Modeling Complex Decision Boundaries Through Oblique Tree Ensembles. The System Is Supported By An Efficient LMDB Database For Fast Data Access And A Tkinter-based Interface For Real-time Interaction. It Predicts Churn Risk Scores And Complaint Status With Improved Accuracy And Interpretability. By Enabling Early Identification Of Churn Patterns And Efficient Large-scale Processing, The Framework Supports Proactive Decision-making And Enhances Overall Service Quality.

Published:

24-4-1-2026

Issue:

Vol. 26 No. 4-1 (2026)

Page Nos:

790-804

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite