Alzheimer's disease (AD) is one of the major conditions suffered by elderly people, one early symptom of which is disorientation. This paper aims to develop and evaluate algorithms to detect the movement patterns of elderly people in order to realize whether they are lost. We then make use of this algorithm to develop a diagnostic tool for cog-nitively impaired elderly people. In this paper, five healthy people are asked to simulate AD subjects in order to build the database with two different labels: (1) normal controls (NCs), and (2) AD subjects. A navigation tool is therefore developed to detect whether the subjects get lost during the well-designed walking navigation experiments. To begin with, we develop this navigation diagnostic tool based on an Android smartphone with a simple graphic user interface, since most elderly people cannot handle too much information at the same time. This diagnostic tool contains four sensors: (1) accelerometer, (2) gyroscope, (3) magnetometer, and (4) global navigation satellite system receiver, in order to calculate the angle of the direction in which the subjects should head. Data on the subjects' acceleration during the experiments is also collected to help analyze the elderly people's cognitive and navigation abilities through signal processing. The raw data collected from the accelerometer is extracted into informative features with a 30-second sliding window. The classification model used to classify whether the subject is an AD patient is the one-nearest-neighbor algorithm. Since the measured behavior is a temporal sequence, an elastic distance measurement, called dynamic time warping distance, is applied for the one-nearest-neighbor algorithm. The experimental results show the effectiveness of this proposed navigation tool, with a classification accuracy of 90% using one-nearest-neighbor algorithm with the mean of z-axis acceleration, and the standard deviation of the resultant acceleration as extracted features.