Improving Basal Insulin Prediction Through Outlier Filtering and Fuzzy Validation for Risk-Free Insulin Pump Therapy

Abstract

Diabetes has become more common in adults and even younger patients due to lifestyle changes. The advent of interventions for self and remote healthcare in a pervasive healthcare context has led to the use of insulin pumps to control blood sugar levels. The insulin pump operation needs the manual configuration of basal insulin levels, which brings the risk of hypoglycemia or hyperglycemia. The risk can be prevented by predicting and validating the safety of basal insulin before dosage, while also considering various dosage levels, which is not done in existing systems. Hence, heuristic data analysis is performed using anomaly detection over the data, removing outliers, and predicting basal insulin. The predicted value is validated with a fuzzy inference system. On deleting outliers, the proposed system gets the best mean absolute error (MAE) of 0.775 using the isolation forest technique. The basal insulin is projected using the methods with the lowest MAE (Long Short-Term Memory, recurrent neural network, and gated recurrent unit [GRU]), with GRU having 2.479 MAE and the best forecasted value. A fuzzy inference system for insulin pumps was modeled with various rule bases for low, moderate, and high dosage levels. The results of this study are evaluated using risk scores and explainable artificial intelligence concepts. This work will enable developers to create risk-free models that can adjust treatment in a number of ways, including how postprandial glucose patterns change, how stable overnight glucose levels are, how long bolus insulin takes to work, when and how to respond to hypoglycemia, and how different lifestyle activities affect treatment.


Received: 13 September 2025 | Revised: 17 December 2025 | Accepted: 6 February 2026



Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.



Data Availability Statement

The data that support the findings of this study are openly available at https://archive.ics.uci.edu/dataset/34/diabetes. Further, for enhanced validation, the model was tested on https://webpages.charlotte.edu/rbunescu/data/ohiot1dm/OhioT1DM-dataset.html, which is available on request.


Author Contribution Statement
Jeyalakshmi Jeyabalan: Conceptualization, Methodology, Investigation, Writing – review & editing, Supervision, Project administration. Karthikeyan Suresh: Software, Validation, Formal analysis, Resources, Data curation, Writing – original draft, Visualization.