<p>Outlier detection aims to identify abnormal objects in data distributions and plays a vital role in improving data quality. Although semi-supervised outlier detection methods have shown effectiveness in improving detection performance by utilizing limited labeled information, existing methods still exhibit deficiencies in the utilization of local density information and the modeling of system disorder, which further restrict their detection performance in complex data distribution scenarios. To address these issues, this paper proposes a semi-supervised outlier detection algorithm based on density feature and fuzzy outlier factor (DFFOF). Specifically, the local density of each sample within an extended neighborhood is extracted as a density feature and fused with the original feature space to construct an augmented feature space, thereby enhancing the representation capability of local structural information. Then, a fuzzy decision system is constructed using a small number of labeled samples to guide the calculation of approximate discernibility, so as to evaluate the capability of features to support decision classification. Furthermore, fuzzy entropy differences are calculated to evaluate the outlier degree of samples on each feature from the perspective of disorder. Finally, the above metrics are integrated in the augmented feature space to construct fuzzy outlier factors for outlier detection. Experimental results on synthetic and real-world datasets demonstrate the effectiveness and superiority of DFFOF in outlier detection tasks.</p>