Nowadays emerging sleep-tracking technologies such as Fibit make it possible for individuals to collect personal sleep data. However, people find it difficult to gain insights from these data without proper analysis. The objective of this study was to investigate the possibility of establishing a sleep analysis approach that helps people detect their unusual sleep pattern by considering their own sleep baselines instead of the population average. The proposed approach was consisted of two steps. In the first step, the dimension of time series sleep data was reduced using permutation entropy. Following that, univariate outlier detection techniques were applied to detect unusual sleep patterns. We tested our approach on a real sleep tracking data set consisting of 35 days of time series data tracked using a Fitbit Charge HR. Depending on the univariate outlier detection technique used, the identified unusual sleep differed. We found that permutation entropy of a sleep time series was strongly correlated to the time that the user went to bed and weekly correlated to minutes asleep, but was not correlated to minutes awake, awakening count and sleep efficiency. Based on the analysis results, we pointed out the directions for future study on personal sleep data analysis.