Introduction : With the increase of sedentary style of life and ageing, quantifying the amount of movement during daily life becomes crucial. For monitoring daily life activities, inertial measurement units (IMU), a combination of miniature tri-axial angular rate and gravity sensors, are promising tool. To this purpose, different studies proposed to apply machine learning methods to the data provided by single or multiple sensors [1] attached to the various segments of the body [2]. The aim of this study was to develop and test amethod to identify some of the daily life activities simply using a single IMU attached at the waist level on a body side. Materials and methods : Data were acquired using a single IMU (FreeSense, Sensorize ® ) featuring a tri-axial accelerometer and two bi-axial gyroscopes (sampling frequency 50 Hz) placed at the waist level on the right body side. Training data set was acquired on ten subjects (four females, 31±4 yrs) who performed the following dynamic activities: walking, walking up/down stairs, turning, sit-to-stand, and stand-to-sit. Static activities performed were: standing, sitting. The subjects were guided by an examiner, who identified the timing of each activity, throughout the 5min long acquisitions. Ten acquisitions were made for each subject. The distinction between dynamic and static activities was accomplished with the method that was presented in a previous study [3]. Then, two different approaches were used for each activity type. Static activities were distinguished based on the relative angle with respect to the gravitational acceleration direction. For the dynamic activities a neural classifier was used. Distinguishing static and dynamic activities allows the neural classifier to focus on fewer types of activities aimed at increasing its accuracy. Multilayered feed-forward artificial neural network (ANN) with the resilient backpropagation (RPROP) learning algorithm was used to train the neural classifier. The ANN presented two hidden layers and a hyperbolic tangent sigmoid transfer function was used as transfer function for each layers. The signal inspection is performed using a sliding window approach, with window size 128 samples (2.56 s) and 64 samples (1.28 s) overlapping between each consecutive window. For each window, features characterizing the signal were extracted from the IMU signals. The selected features were: (1) the mean value of the signal within the window interval, (2) the standard deviation of the window interval, (3) the energy content (sum of squared FFT component magnitudes divided by the window length), (4) correlation between signals in the interval of window. The correlation was useful for differentiating among activities in which most of the movement is along a sensing axis. For example, walking and running feature the largest displacement in one of the sensor directions, whereas stair climbing involves a displacement in a combination of two sensor directions. Effectiveness of these features was demonstrated in prior work [1,2]. Thirty-three features for each window were extracted: six features (three for the acceleration and three for the angular velocities) for mean, standard deviation and energy and, fifteen features for correlation were calculated. 75% of data acquired were used for training, 15% for validation and 10% of the acquired data is used for testing the method. Results : Average recognition and accuracy rates for the whole process were 95.21% and 94.87%. The recognition rate column shows the percentage of activities classified correctly. Accuracy rate column shows the percentage of predicted activities that are classified correctly. Discussion : In this study, amethod to classify the daily life activities using an artificial neural network using a single IMU attached to the waist was proposed. The average accuracy rate in this study was higher than a similar method proposed previously [1] which was 90.61%. This might be due to the fact that in the referenced study only accelerometer signals were used, whereas in this study gyroscope signals were also used.