A quality prediction method for weight lifting activity

• Published in: Michael Faraday IET International Summit 2015 p. 95
• Main Authors: Chatterjee, S, Chakraborty, R, Dey, N, Hore, S
• Format: Conference Proceeding
• Language: English
• Published: Stevenage, UK IET 2015; The Institution of Engineering & Technology
• Subjects: Activity recognition; Bayesian analysis; Errors; Feature recognition; Hoisting; Humanities computing; Mathematical models; Neural computing techniques; Other topics in statistics; Training; sports activity recognition; false positive rate; TP rate; weight lifting activity; sports athlete; FP rate; F-Measure; quality prediction method; random forest; performance measures; relative absolute error; perfect execution; root mean squared error; multilayer perceptrons; mean absolute error; pattern classification; multilayer feedforward network; HNB classifiers; true positive rate; machine based feedback; kappa statistic; RMSE; MAE; RBFNN; health condition; hidden Naive Bayes; RAE; Bayes methods; sport; MLP-FFN; immense interest
• ISBN: 9781785611865; 1785611860
• DOI: 10.1049/cp.2015.1691

Activity recognition has found immense interest in the field of sports activity recognition in recent time. The application has found extensive utility in giving machine based feedback on how well the performance in training is by any sports athlete. The present work proposes a model to predict the quality of training along with the mistakes which can have a severe effect on performance of the athlete or health condition of any subject. The dataset contains data of perfect execution of weight lifting activity and the same with four common mistakes. 34 features have been selected and used to train and test the proposed model. A multi-layer feed-forward network (MLP-FFN), RBFNN, Random Forest and Hidden Nai¨ve Bayes (HNB) classifiers are employed to determine the objective. The assessment of used methods has been done by observing different performance measures such as Kappa statistic, Mean absolute error (MAE), Root mean squared error (RMSE), Relative absolute error (RAE), True positive (TP) Rate, False positive (FP) Rate and F-Measure. The experimental results have shown almost perfect classification using MLP-FFN and satisfactory results for all the proposed models.