Table 1.

Title, goal of related work, number of horses in the dataset, sensors and methods proposed in this paper with related work. What makes our research unique is the accurate detection of a the high number of relevant jumping and dressage training activities by automatically extracting features with a CNN using accelerometer data of two legs. NN = Neural Network, DT = Decision Trees, k-NN = k-Nearest Neighbors, NB = Naive Bayes, CNN = Convolutional Neural Network, LDA = linear discriminant analysis, QDA = quadratic discriminant analysis, RF = Random Forest, SVM = Support Vector Machine, LSTM = Long short-term memory, BT = Boosted Trees, GPR = Gaussian Progress Regression.

Paper	Goal	Number of Horses	Sensors	Classification Approach
[5]	Detection of walk, trot and canter	2	Accelerometer	NN, DT, k-NN and NB
[6]	Standing, grazing and ambulating	6	Accelerometer	Threshold based
[11,14]	Detection of stand, walk, trot, canter, roll, paw, flank-watching	6	Accelerometer	CNN
[15]	Detection of stand, walk, trot and canter	20	Accelerometer	Threshold based
[7]	Walk, trot, left canter, right canter, tölt, pace, trocha and paso fino	120	Accelerometer + gyroscope	LDA, QDA, DT, RF, SVM, NN and LSTM
[16]	Estimation of speed in canter	58	Accelerometer + gyroscope	SVM
[17]	Estimation of speed in walk, trot, tölt, pace and canter	40	Accelerometer + gyroscope	SVM, DT, RF, BT, GPR
[18]	Presence/absence and degree of lameness	175	Camera	NN
[4,9,19]	Detection of collected, working, medium and extended pace	6	Camera	Threshold based
[10]	Detection of trot, piaffe and passage	10	Camera	DA
[20]	Gait analysis	35	Strain gauge	NN
[21]	Hoof wall deformation to determine ground reaction forces	1	Strain gauge	NN
[22]	Prediction of load in long bone	9	Strain gauge	NN
[23]	Load-displacement in long bone	13	Strain gauge	NN
This paper	6 jumping and 25 dressage training activities	14	Accelerometer	Hybrid CNN