The potential of machine learning for weather index insurance

Abstract

Abstract. Weather index insurance is an innovative tool in risk transfer for disasters induced by natural hazards. This paper proposes a methodology that uses machine learning algorithms for the identification of extreme flood and drought events aimed at reducing the basis risk connected to this kind of insurance mechanism. The model types selected for this study were the neural network and the support vector machine, vastly adopted for classification problems, which were built exploring thousands of possible configurations based on the combination of different model parameters. The models were developed and tested in the Dominican Republic context, based on data from multiple sources covering a time period between 2000 and 2019. Using rainfall and soil moisture data, the machine learning algorithms provided a strong improvement when compared to logistic regression models, used as a baseline for both hazards. Furthermore, increasing the amount of information provided during the training of the models proved to be beneficial to the performances, increasing their classification accuracy and confirming the ability of these algorithms to exploit big data and their potential for application within index insurance products.

The Role of Data-Driven Methodologies in Weather Index Insurance

“…Figure 3 summarizes the methodology used in this study which is adapted from two recent methodologies: Cesarini et al [24] and Figueiredo et al [31]. According to [27], the performance of a model can be affected by selecting a "bad algorithm" or "bad data"; in addition, for most machine learning algorithms, a considerable amount of data is needed for them to work well [27].…”

“…For the specific case developed, selection criteria were implemented to build a representative database of the problem. Additionally, some of the selected predictors were proven to be significant in previously developed research, such as the database selected by Cesarini et al [24].…”

“…Second, we want to compare a linear and interpretable model with a highly non-linear black-box model. Therefore, we will establish whether a machine learning model, such as the one recently used in the literature to model weather indices and loss [24], can be employed in a productive region in Colombia to establish if there is a relationship between weather indices and the observable losses due to floods. We used a logistic regression model as a reference point to evaluate the performance of a more complex model, ANN.…”

“…This paper contributes to the state of the art as the application of machine learning methods has been limited in most low-income countries. This limitation has been shown even though machine learning methods have demonstrated their potential as non-linear and flexible methods for estimating performance-index relationships [24]. Furthermore, we believe that the research in index insurance programs should be promoted in low-income economies, especially in countries with social conflicts, as is the case of Colombia.…”

“…The implementation of decision trees has proved useful in identifying important weather indices from satellite data [23] and for capturing non-linear relationships between different crop classes and low levels of precipitation, and evapotranspiration [5]. Additionally, Cesarini et al [24] compare the performance of neural networks and support vector machines for identifying floods and droughts from satellite data, demonstrating the superiority of both methods over a logistic regression model [24]. There are other types of studies involving machine learning algorithms in weather index insurance.…”