Jan 2023 – Jan 2024 · Granted Patent

Cardiac Risk Prediction with a Hybrid Fuzzy Inference System

A membership function that keeps the interpretability of fuzzy logic without flattening the shape of real clinical risk. Granted as an Indian patent.

Patent granted: "System and Method for Predicting a State of a Heart Using a Hybrid Fuzzy Membership Model" — Indian Patent No. 601763 (Application No. 202441004228), granted 8 September 2026

Results

  • 56% average improvement in precision over a standard fuzzy membership model — 0.61 vs. 0.39 in direct comparison
  • 59% average improvement in F1-score — 0.43 vs. 0.27 in direct comparison
  • Validated across three distinct datasets spanning general cardiac risk assessment, specific anomaly detection, and patient profiling, to confirm the approach generalizes rather than overfitting to one dataset
  • Binary Brain Storm Optimization (BBSO) for feature selection reduced model complexity while improving diagnostic performance, rather than trading one for the other
  • Granted as Indian Patent No. 601763 (Application No. 202441004228, filed 2024, granted September 2026), co-invented with Siva Shanmugam G, Anbarasi M, and Anusha Seshadri (VIT)

Stack

  • Python
  • Fuzzy Inference Systems
  • Binary Brain Storm Optimization (BBSO)
  • scikit-learn

The granted patent certificate — Patent No. 601763, granted 8 September 2026.

FIG. 1 from the granted patent — the full pipeline, end to end.

InteractiveHow the hybrid membership function was built

The trapezoid

A trapezoid rises linearly, holds flat at 1 across its plateau, then falls linearly back to 0. It's the standard building block for fuzzy sets — cheap to compute, and its corners give you a clean, readable boundary: below this point, not in the set at all; above this point, fully in it.

The problem: every value across the plateau reads as an identical "1." A value just inside the boundary and a value dead center are indistinguishable — the function has no gradient once you're "in."

← Back to all work