id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
work_vs53ce2hjzfbha3l3pbs4j24be	Violeta Mirchevska	Combining domain knowledge and machine learning for robust fall detection	2013	22	.pdf	application/pdf	8532	882	64	Third, the classifier is adapted online based on user feedback using the Markov decision process. Keywords: prior domain knowledge, inductive machine learning, ambient intelligence, fall detection fall detection domain, an expert may specify that if an elderly person is lying or sitting on the ground for In the fall detection example, the starting rule-based classifier contained the following rule types: 1. IF falling activity within T1fall seconds AND the user was lying/sitting on the ground P1activity% of T1activity the last rule of the fall detection classifier, presented an issue for the expert, as the values may be rule type requires detecting falling activity and the user to be immovable and lie/sit on the ground to detect a fall, whereas the fourth rule type requires only the user to lie/sit on the ground. initialized to zero for all states and MDPR.current is set to the rule's values in the refined classifier (Figure 	./cache/work_vs53ce2hjzfbha3l3pbs4j24be.pdf	./txt/work_vs53ce2hjzfbha3l3pbs4j24be.txt