Innovative use of machine learning to forecast crop disease risk

“There is added value in combining the algorithms in an ensemble to provide a more accurate and robust forecasting tool that can be tailored to produce region-specific alerts. The techniques used can easily be applied to outbreak data from other crop diseases to derive tools to help farmers and land managers make the best decisions.”

Crop diseases can generate destructive outbreaks that have the potential to threaten global food security, which is why it is fundamental to have reliable data promptly available from disease surveillance programs and outbreak investigations. In many cases, however, only information on outbreaks is collected and data from surrounding healthy crops is omitted. Use of such data to develop models that can forecast risk/no-risk of disease is therefore problematic, as information relating to the no-risk status of healthy crops is missing.

FindOUT, a new application developed by James Hutton Institute bioinformatician Dr Peter Skelsey, employs – for the first time – machine-learning ‘anomaly detection algorithms’ to forecast the risk of crop disease, greatly increasing the accuracy of forecasts. The desktop app enables end-users to train and test their own anomaly detection crop disease forecasting tool using their own data.

In a new study published in the journal Phytopathology, edited by the American Phytopathological Society, Dr Skelsey describes a novel approach to disease forecasting based on data comprised of outbreaks only.

This was done in two steps: in the training phase the algorithms were used to learn the envelope of weather conditions most associated with historic crop disease outbreaks, while a second phase saw the algorithms being used to process historic outbreak events.

Five different anomaly-detection algorithms were compared according to their accuracy in forecasting outbreaks: robust covariance, one-class k-means, Gaussian mixture model, kernel density estimator, and one-class support vector machine.

Dr Skelsey used a case study of potato late blight survey data from across Great Britain as proof-of-concept; results showed that the Gaussian mixture model had the highest forecast accuracy at 97.0%, followed by one-class k-means at 96.9%.

Dr Skelsey commented: “There is added value in combining the algorithms in an ensemble to provide a more accurate and robust forecasting tool that can be tailored to produce region-specific alerts.

“The techniques used can easily be applied to outbreak data from other crop diseases to derive tools to help farmers and land managers make the best decisions.”

There has been great interest in this work, and Dr Skelsey recently received additional funding to further develop and refine the desktop app into a commercial / licensed application.

The open-access paper Forecasting risk of crop disease with anomaly detection algorithms is available online, and findOUT can be downloaded at https://github.com/pskelsey/findOUT.