With this effort, we want to provide a pathway from raw behavioral data to a behavioral model that can identify online identity theft effectively, quickly, and robustly. We zero in on this problem in OSNs, where users often keep composite records of their behaviors, including offline check-ins and online user-generated content (UGC), which is multimodal and of poor quality. We verify that various record dimensions have a complementing impact when modeling users' behavioral patterns, which is an informative discovery. We recommend a combined (rather than fused) model to record aspects of a user's composite behavior that occur both online and offline in order to fully take use of this complimentary impact. We test the suggested combined model on two real-world datasets, Foursquare and Yelp, and compare it to both standard models and their combined model. Our model beats the current ones, according on the experimental data. The area under the receiver operating characteristic curve (AUC) values were 0.956 for Foursquare and 0.947 for Yelp, respectively. In example, with a disturbance rate (false-positive rate) of less than 1%, the recall (true positive rate) may reach as high as 72.2% in Yelp and 65.3% in Foursquare.