Wandering behaviour is a neurological disorder experienced by elderly people with cognitive impairment, affecting normal brain functioning of the patients. In extreme cases, the patients either completely or partially loses control to perform activities of daily livings.This paper presents privacy-preserving, easily deployable wandering behaviour detection system in dementia patients by identifying their day to day activities and detecting wandering behaviour when they leave area of interest. In this context, we have used low-cost small wireless devices such as off-the-shelf Wi-Fi router and network interface card operating at 2.4 GHz. The Wi-Fi data collected is converted into scalograms using continous wavelet transform. An unsupervised deep learning algorithm
namely Autoencoder is used to classify scalgorams to identify critical event such as wandering behaviour. Modified Logistic and Dynamic Newton Leipnik Maps are used to encrypt the health record of patients to preserve the privacy. The
proposed system provided high classification of accuracy 94.2%. Furthermore, the proposed system is robust in terms of security and has been tested on a number of parameters such as correlation coefficient (0.0001), number of pixel change
rate (99.9%), unified average change intensity (33.2), energy (0.01) and contrast (10).