2022

Rautiainen, Ilkka; Kauppi, Jukka-Pekka; Ruohonen, Toni; Karhu, Eero; Lukkarinen, Keijo; Äyrämö, Sami

Overweight, obesity and diseases associated with them have been increasing rapidly during the last few decades. The early detection of obesity risk is the key to preventive actions. This task can be supported by machine learning-based prediction models that reliably predict future overweight/obesity status based on early childhood data. The case study presented here employs predictive modeling and height/weight data collected by the health care system of Äänekoski town, Finland. The study utilizes nine existing study designs carefully selected based on a recent literature review. For each individual in the data, the BMI growth curves were resampled at 30-d intervals using the linear interpolation technique. This time series data is then utilized to form several predictive models using logistic regression, support vector machine, and a decision tree. Prediction accuracy is comparable to existing studies, and in some cases, even better. The best model, trained by the SVM method on the Finnish data, obtained an F1-score of 0.73. The results suggest that the Finnish data may contain strong dependencies that can be utilized in building the models. However, more versatile information from the early years of childhood is most likely needed to further optimize the models.