1. Title of the paper Evolving Classification Rules for Predicting Hypoglycemia Events 2. Authors Marina De La Cruz^1 Carlos Cervig˘n^1 Jorge Alvarado^1 Marta Botella-Serrano^2 J Ignacio Hidalgo^1 ^1 Facultad de Inform tica, Universidad Complutense de Madrid, Calle del Profesor Jos‚ Garcˇa Santesmases 9, 28040 Madrid absys@ucm.es, +34913947537 ^2 Hospital Universiatario Principe de Asturias Av. Principal de la Universidad, s/n, 28805 Alcal  de Henares, Madrid, marta.botella@salud.madrid.org 3. Corresponding author J Ignacio Hidalgo hidalgo@ucm.es 4. Abstract People with diabetes have to properly manage their blood glucose levels in order to avoid acute complications. This is a difficult task, and an accurate and timely prediction may be of vital importance, especially of extreme values. Perhaps one of the main concerns of people with diabetes is to suffer a hypoglycemia (low value) event and moreover, that the event will be prolonged in time. It is crucial to predict events of hyperglycemia (high value) and hypoglycemia that may cause health damages in the short term and potential permanent damages in the long term. The aim of this paper is to describe our research on predicting hypoglycemia events using Dynamic structured Grammatical Evolution. Our proposal gives white box models induced by a grammar based on if-then-else conditions. We trained and tested our system with real data collected from 5 different diabetic patients, producing 30-minute predictions with excellent results. 5. Criteria satisfied A, B, D, F, G 6. Justification of criteria satisfied Predictors of hypoglycemia events were evolved to generate alarm systems for persons with Diabetes. Those Predictors imitate the way a human would reason, i.e., by if- then-else statements. We improve the quality that a human could reach by incorporating information of variables measured using smartwatches, that are not used by humans. There is a need for reliable predictive models of blood glucose low levels (hypoglycemia) for People with Diabetes. The prediction of a future hypoglycemic episode can lead the patient to take measures to remedy it before it happens and to take better decisions in the future to avoid acute complications. An important aspect is that the symptoms of hypoglycemia vary depending on the person but usually encompass: shaking, sweating, hunger, fast or irregular heartbeat, numbness in extremities, etc... These symptoms can be easily confused whilst exercising and a person with diabetes might not be aware that she/he is suffering an hypoglycemia event. Human predictions are made using only variables such as the glucose at the time of prediction. We train classification models by means of Structured Grammatical Evolution (SGE). We present a method to obtain White-box models composed by a set of {if-then-else} conditions. Evolution allows to evaluate information of a set of physiological variables during the last two hours previous to the prediction time. In particular, we include the values of glucose levels measured by a Continuous Glucose Monitoring System (CGM), and heart rate information, number of steps and calories burned (these last three obtained by wearing a smartwatch). Another thing to take into account is that the body can get used to being in a state of hypoglycemia, this means that the more hypoglycemic episodes a patient suffers the less they will feel the initial symptoms until their blood sugar drops lower and lower, which is not ideal when trying to recognize the state if the patient is not actively looking at their current blood sugar levels. One of the main advantages of SGE and DSGE models is that they generate interpretable expressions, i.e the expressions explicitly present the variables used and how they interact with each other. We investigated the performance of two types of Structured Grammatical Evolution for the prediction of the classification of future glucose values on a short prediction horizons (30-120 minutes), with input data obtained automatically (through a smartwatch and a CGM). We have also tested the creation of a general model that can be applied to all patients and how this model compares with the individual ones. Evolution provides several interesting results: From the results obtained, we can discern that the best results are obtained with the individualized models, but the general model can yield similar TPR and TNR results on short time horizons. When performing the test on each patient we observed that some follow widely different patterns. This effect can be helped using clusters, and it is recommended when the number of patients increases, since the more patients we add the more probable that some of them are not considered in the final model obtained. Nonetheless, general models are not recommended for a longer time frame unless we can find a set of patients whose glucose development is very similar. As previously stated, the resulting models obtained are an if-then-else statement with a series of conditions that include the input variables and some constants. As the models obtained are made up of numeric, relational, and logical operations between variables, they can be explained and studied after being obtained, edited, and retested. This freedom opens a lot of possibilities for tweaking good solutions after being obtained to fit better the problem or to study those cases where the model has failed and why it has failed. In terms of the criteria: (A) The result would qualify today as a patentable new invention. The predictors are now incorporated to a mobile application which generates alarms for persons with diabetes. (B) The result is equal to or better than a result that was accepted as a new scientific result at the time when it was published in a peer-reviewed scientific journal. Due to the importance of hypoglycemic prediction, a variety of techniques have been used to forecast future hypoglycemic events with different types of input data. To our knowledge, this is the best accuracy so far using classification. (D) Hypoglycemia prediction is publishable in its own right as explained above, independent of the fact that the result was mechanically created. (F) Due to the importance of hypoglycemic prediction, a variety of techniques have been used to forecast future hypoglycemic events with different types of input data. (G) The result solves a problem of indisputable difficulty in its field. 7. Citation M. De La Cruz, C. Cervig˘n, J. Alvarado, M. Botella-Serrano and J. I. Hidalgo, "Evolving Classification Rules for Predicting Hypoglycemia Events," 2022 IEEE Congress on Evolutionary Computation (CEC), Padua, Italy, 2022, pp. 1-8, doi: 10.1109/CEC55065.2022.9870380. 8. Prize money, if any, will be divided as follows: Marina de la Cruz:30% J. Ignacio Hidalgo:30% Carlos Cervig˘n:30% Jorge Alvarado: 5% Marta Botella: 5% 9. Why this entry is the best As previously stated, the resulting models obtained are an if-then-else statement with a series of conditions that include the input variables and some constants. As the models obtained are made up of numeric, relational, and logical operations between variables, they can be explained and studied after being obtained, edited, and retested. This freedom opens a lot of possibilities for tweaking good solutions after being obtained to fit better the problem or to study those cases where the model has failed and why it has failed. Grammatical evolution came up with better models. Experimental results demonstrates that evolutionary models are more precise that human experts could be. We have integrated this technique in a real-world mobile app, capable of generating alarms signals and improving the quality of life of people with diabetes. 10. Type of EC used Grammatical Evolution 11. Date of Publication September 6, 2022