Deep Learning at the Interface of Agricultural Insurance Risk and Spatio-Temporal Uncertainty in Weather Extremes

Ghahari, Azar; Newlands, Nathaniel K.; Lyubchich, Vyacheslav; Gel, Yulia R.

doi:10.1080/10920277.2019.1633928

Cited by 21 publications

(1 citation statement)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The advantages of incorporating nonlinear assumptions and blended predictors, which involve the combination of measured variables with average or extreme events, are numerous. Findings show the potential of deep learning for prediction in insurance (Ghahari et al, 2019).…”

Section: Introductionmentioning

confidence: 85%

Predictive Analytics and Risk Management of Property Insurance

Amirhosseini,

Vanani,

Taghavifard

et al. 2023

Preprint

View full text Add to dashboard Cite

Claim prediction plays a leading role in risk management for insurance companies. This research proposes a risk management model by evaluating property policy's risk and classifying risk levels to reduce uncertainty for decision-makers. The approach consists of three phases. Phase I is claims prediction; It integrates three claim predictions: occurrence probability, claim severity, and time of loss; we use machine learning for each prediction. Algorithms have been implemented on about 2 million records of a large insurance company between 2011 and 2022. Results showed that the best algorithm is a deep neural network. In Phase II, risk metric results are calculated to determine the risk level score; we present the prediction results into a quadrant risk-grouped associated measure. Phase III is risk management. The risk level score is linked to the decision-makers action list to keep, reject or transfer the risk of each insurance policy. The final model is a risk management tool combining risk prediction metrics and the risk matrix. To assess the efficacy of this new model, 500 sample records were provided to the risk management system, and its output was compared with that of expert opinion. 72% correct matching indicates the accuracy of the model.

show abstract

Section: Introductionmentioning

confidence: 85%