Human Activity Recognition in AAL Environments Using Random Projections

Table 3

Summary of related works in the HAR domain.

Author	Activities	Sensor data	Features	Feature selection	Classification method	Accuracy
Atallah et al. [39]	Lying down, preparing food, eating and drinking, socialising, reading, dressing, walking, treadmill walking, vacuuming, wiping tables, running, treadmill, running, cycling, sitting down/getting up, and lying down/getting up	Acceleration sensors	Averaged entropy over 3 axes, main FFT frequency (averaged) over 3 axes, energy of the 0.2 Hz window centred around main frequency over total FFT energy (3-axis average), and averaged mean of cross covariance between every 2 axes	ReliefF, Simba, and MRMR	kNN, Bayesian classifier	90%
Bayat et al. [40]	Running, slow walk, fast walk, aerobic dancing, stairs up, and stairs down	Triaxial accelerometer	Mean along -axis, MinMax, STD, and RMS for Am, APF along -axis, -axis, and -axis, VarAPF, STD along -axis, -axis, and -axis, RMS along -axis, -axis, and -axis, correlation between -axis and -axis, and MinMax along -axis, -axis, and -axis	Feature clustering	Multilayer perceptron, SVM, Random Forest, and Logit Boost	81%–91%
Berchtold et al. [41]	Standing, sitting, lying, walking, climbing stairs, cycling, and being stationary	Accelerometer	Variance, mean	None	Fuzzy inference	97.3%
Capela et al. [17]	Sitting, standing, and lying; ramp up and ramp down; stairs up and stairs down; transition between activities	Linear acceleration, gravity, and velocity sensors	Range, mean, standard deviation, kurtosis, moving average, covariance matrix, skewness, zero cross rate, and mean cross rate	None	Naïve-Bayes, Support Vector Machine, and j48 decision tree	97%
Gupta and Dallas [30]	Jumping, running, walking, sitting, sitting-to-standing, and standing-to-kneeling	Triaxial accelerometer	Energy, entropy, mean, variance, mean trend, windowed mean difference, variance trend, windowed variance difference, detrended fluctuation analysis coefficients, --energy, and max. difference acceleration	ReliefF, SFFS	kNN, Naive Bayes	98%
Henpraserttae et al. [42]	Sitting, lying, standing, and walking	Accelerometer	Mean and standard deviation	None	Rules and threshold based classification	90%
Hoque and Stankovic [43]	Leaving house, using toilet, taking shower, sleeping, preparing breakfast, preparing dinner, getting snack, getting drink, using washing machine, and using dishwasher	Location sensors (open/closed)	Magnitude	None	Custom clustering method	64.5%–89.9%
Iso and Yamazaki [44]	Walking, running, stairs up/down, and fast walking	Accelerometer	Wavelet components, periodograms, and information entropy	None	Bayesian probabilities	80%
Kose et al. [45]	Walking, running, biking, sitting, and standing	Accelerometer	Min., max., average, variance, FFT coefficients, and autocorrelation	None	Clustered kNN	95.2%–97.5%
Kwapisz et al. [46]	Walking, jogging, stairs up/down, sitting, and standing	Accelerometer	Mean, std. dev., average absolute difference, average resultant acceleration, time between peaks, and binned distribution	None	Decision tree, logistic regression, and MNN	91.7%
Lane et al. [47]	Driving, being stationary, running, and walking	GPS, accelerometer, and microphone	Mean, variance	None	Naïve-Bayes	85–98%
Lee and Cho [48]	Standing, walking, running, stairs up/down, shopping, and taking bus	Accelerometer	-, -, and -axes acceleration values	None	Hierarchical HMM	70%–90%
Mannini and Sabatini [49]	Walking, walking carrying items, sitting & relaxing, working on computer, standing still, eating or drinking, watching TV, reading, running, bicycling, stretching, strength training, scrubbing, vacuuming, folding laundry, lying down and relaxing, brushing teeth, climbing stairs, riding elevator, and riding escalator	Acceleration sensors	DC component, energy, frequency-domain entropy, and correlation coefficients	SFFS (Pudil algorithm)	Continuous emissions, Hidden Markov Model	99.1%
Mathie et al. [37]	Various human movements, including resting, walking, and falling	Triaxial acceleration sensor	Integrated area under curve	None	Binary decision tree	97.7% (sensitivity) 98.7% (specificity)
Maurer et al. [50]	Walking, standing, sitting, running, and ascending and descending the stairs	Multiple sensors	Mean, root mean square, standard deviation, variance, mean absolute deviation, cumulative histogram, th percentiles, interquartile range, zero crossing rate, mean crossing rate, and sq. length of ,	Correlation-based Feature Selection (CFS)	Decision trees (C4.5 algorithm), -Nearest Neighbor, Naïve-Bayes, and Bayes Net	80%–92%
Miluzzo et al. [51]	Sitting, standing, walking, and running	Accelerometer, GPS, and audio	DFT, FFT features, mean, std. dev. and number of peaks per unit, and time deviation of DFT power	None	Decision tree	79%
Pärkkä et al. [52]	Lying down, rowing, ex-biking, sitting/standing, running, and Nordic walking	GPS, audio, altitude, EKG, accelerometer, compass, humidity, light, temperature, heart rate, pulse, respiratory effort, and skin resistance	Peak frequency of up-down chest acceleration, median of up-down chest acceleration, peak power of up-down chest acceleration, variance of back-forth chest acceleration, sum of variances of 3D wrist acceleration, and power ratio of frequency bands 1–1.5 Hz and 0.2–5 Hz measured from chest magnetometer	Heuristic	Decision tree	86%
Saponas et al. [53]	Walking, jogging	Accelerometer	124 features: Nike + iPod Packet Payload, magnitude (mean, std. dev., min., max., and min. minus max.), frequency (energy in each of the first 10 frequency components of DFT, energy in each band of 10 frequency components, largest frequency component, and index of the largest frequency component)	None	Naïve-Bayesian Network	97.4% (within-person), 99.48% (cross-person)
Siirtola and Röning [54]	Walking, running, cycling, driving, sitting, and standing	Accelerometer	Magnitude, std., mean, min., max., percentiles (10, 25, 50, 75, and 90), and sum and square sum of observations above/below percentile (5, 10, 25, 75, 90, and 95) of magnitude acceleration and square sum of   &	None	Decision tree + kNN/QDA	95%
Sohn et al. [55]	Walking, driving, and dwelling	GPS	Spearman rank correlation, variance, and mean Euclidean distance over a window of measurements	None	Logistic regression	85%
Yang [56]	Sitting, Standing, walking, running, driving, and bicycling	Accelerometer	Mean, std., zero crossing rate, 75th percentile, interquartile, spectrum centroid, entropy, and cross-correlation	None	Decision tree, Naïve-Bayes, kNN, and SVM	90%
Zhu and Sheng [57]	Sitting, standing, lying, walking, sitting-to-standing, standing-to-sitting, lying-to-sitting, and sitting-to-lying	3D acceleration	Mean, variance	None	Neural network ensemble	67%–98%