A hybrid approach to model randomness and fuzziness
summary
Prototype integrates Dempster Shafer evidence fusion with logistic regression priors to convert point probabilities into belief intervals, explicitly modeling both randomness and fuzziness. Implementation generates per-feature mass assignments, normalizes by conflict, and reclassifies; tested on UCI datasets including email spam and a 23-reader cardiac SPECT set. Results show modest but consistent gains in precision and true positives, sometimes worse false positives and unchanged AUC; scalability and multi-class extension remain open.
AI-written summary · about
chapters (18)
- 0:00Hybrid model randomness and fuzziness
- 1:04Motivation for uncertainty modeling
- 2:08Philosophy of uncertainty and fuzzy logic
- 3:13Randomness and fuzziness definitions
- 5:04Dempster-Shafer theory selection
- 6:50History of Dempster-Shafer theory
- 7:54Dempster-Shafer examples and belief
- 13:17Project objectives for fuzziness prediction
- 15:00Hybrid logistic regression with Dempster-Shafer
- 17:53Merging beliefs in binary classification
- 20:42Modified Dempster-Shafer framework
- 23:03Belief, plausibility, and uncertainty
- 26:15Algorithm training framework
- 28:43UCI datasets and spam testing
- 30:50Cardiac imaging results
- 33:53Future research directions
- 36:09Handling imbalanced datasets
- 38:59Conclusion and publications
related talks
10mHallucinations: Harvesting Uncertainty
51mRandom Projections for Probabilistic Inference
43mProbabilistic Programming in the Real World
52mCredal Models for Uncertainty and Logic Treatment
31mBias Meets Bayes: A Bayesian Perspective on Improving Model Fairness
20mProbabilistic Programming with Stochastic Probabilities
01h 27mFlexible Statistical Modeling
45mTBD