Development and Evaluation of Methodology for Personal Recommendations Applicable in Connected Health

In this paper, a personal recommendation system of outdoor physical activities using solely user’s history data and without application of collaborative filtering algorithms is proposed and evaluated. The methodology proposed contains four phases: data fuzzification, activity usefulness calculation, estimation of most useful activities, activities classification. In the process of classification several data mining techniques were compared such as: decision trees algorithms, decision rules algorithm, Bayes algorithm and support vector machines. The proposed algorithm has been experimentally validated using real dataset collected in a certain period of time from a community of 1000 active users. Recommendations generated by the system were related to weight loss. The results show that our generated recommendations have high accuracy, up to 95%.