Activity Recognition Dataset
----------------------------
(c) 2011, Susanna Pirttikangas, Kaori Fujinami, Jaakko Suutala

Please cite the related article [1], when using the dataset.

Dataset
-------
File "activitydata.txt" contains pre-calculated features from 3D accleration sensors, attached to
subject's necklace, right knee, left wrist, right wrist. From each 3x4 sensor channels mean and 
standard deviation in 0.7 seconds (non-overlapping) window are calculated, procuding 
24-dimensional feature vector at each time step. Initial sapmling rate was 10Hz. Dataset is in 
form n x d+1 (n number of examples, d dimension), where each row presents one feature vector and
each column the value of particular feature, last column being a class label (i.e, the activity 
to be recognized).

Type of each feature and order number are presented below:

From Necklace	Right Knee	Left Wrist	Right Wrist
(1) mean(AccX)	(7) mean(AccX)	(13) mean(AccX)	(19) mean(AccX)
(2) mean(AccY)	(8) mean(AccY)	(14) mean(AccY)	(20) mean(AccY)
(3) mean(AccZ)	(9) mean(AccZ)	(15) mean(AccZ)	(21) mean(AccZ)
(4) std(AccX)	(10) std(AccX)	(16) std(AccX)	(22) std(AccX)
(5) std(AccY)	(11) std(AccY)	(17) std(AccY)	(23) std(AccY)
(6) std(AccZ)	(12) std(AccZ)	(18) std(AccZ)	(24) std(AccZ)

Activity labels and the number of examples in each category are presented below. Dataset contains 13 subjects (P1-P3) and the 
feature vector rows are ordered similarly. For example, if you want to find the starting time of P1's "sit and relax"-activity,
you can calculate it by 55+107+121=283. More details about the dataset and experiments see [1],[2], and [3].

Activity		P1  	P2  	P3  	P4  	P5  	P6  	P7 	P8  	P9  	P10  	P11  	P12  	P13
---------------------------------------------------------------------------------------------------------------------------
1=clean white board	55	78	77	97	98	99	106	42	39	81	24	96	98
2=sit and read news	107	99	103	102	103	100	105	104	104	100	137	95	116
3=stand	  		121	92	94	92	94	90	95	95	91	96	87	90	90
4=sit and relax		96	94	99	97	97	92	104	99	99	94	95	93	95
5=sit and watch TV	121	173	169	170	52	163	183	169	173	163	164	173	169
6=drink			56	74	106	157	48	474	201	63	27	55	48	154	74
7=brush your teeth	93	94	91	94	98	92	98	98	91	90	97	100	97
8=lie down		94	105	37	96	98	77	102	96	93	14	69	100	25
9=vacuum cleaning	218	257	230	193	213	138	253	254	137	168	176	113	185
10=type 		95	95	113	110	101	34	131	101	93	37	95	95	110
11=walk			238	772	349	166	271	113	246	334	112	0	310	378	71
12=walk stairs up	35	36	31	35	30	23	33	38	30	17	44	51	10
13=walk stairs down	31	29	30	29	28	16	30	29	25	34	40	24	9
14=elevator up		32	28	29	30	16	42	24	31	23	54	26	52	26
15=elevator down	19	26	31	29	20	24	20	28	24	27	27	25	24
16=run			15	25	23	8	14	20	29	17	16	0	19	17	18
17=cycle		0	69	114	63	36	74	87	126	59	48	30	86	173



References
----------
[1] Pirttikangas S., Fujinami K. & Nakajima T. (2006) Feature selection and activity recognition from wearable sensors. 
Proc. H.Y. Youn, M. Kim, and H. Morikawa (Eds.):UCS 2006, LNCS 4239, Springer-Verlag Berlin Heidelberg, pp. 516-527.
[2] Suutala J., Pirttikangas S. & Röning J. (2007) Discriminative temporal smoothing for activity recognition from wearable sensors. 
Proc. 4th International Symposium on Ubiquitous Computing Systems (UCS07), Tokyo, Japan, pp. 182-195.
[3] Suutala J. (2012) Learning Discriminative Models from Structured Multi-sensor Data for Human Context Recognition, Doctoral Thesis,
University of Oulu.