Abstract Introduction Smoking lapses after the quit date often lead to full relapse. To inform the development of real time, tailored lapse prevention support, we used observational data from a popular smoking cessation app to develop supervised machine learning algorithms to distinguish lapse from non-lapse reports. Aims and Methods We used data from app users with ≥20 unprompted data entries, which included information about craving severity, mood, activity, social context, and lapse incidence. A series of group-level supervised machine learning algorithms (eg, Random Forest, XGBoost) were trained and tested. Their ability to classify lapses for out-of-sample (1) observations and (2) individuals were evaluated. Next, a series of individual-level and hybrid algorithms were trained and tested. Results Participants (N = 791) provided 37 002 data entries (7.6% lapses). The best-performing group-level algorithm had an area under the receiver operating characteristic curve (AUC) of 0.969 (95% confidence interval [CI] = 0.961 to 0.978). Its ability to classify lapses for out-of-sample individuals ranged from poor to excellent (AUC = 0.482–1.000). Individual-level algorithms could be constructed for 39/791 participants with sufficient data, with a median AUC of 0.938 (range: 0.518–1.000). Hybrid algorithms could be constructed for 184/791 participants and had a median AUC of 0.825 (range: 0.375–1.000). Conclusions Using unprompted app data appeared feasible for constructing a high-performing group-level lapse classification algorithm but its performance was variable when applied to unseen individuals. Algorithms trained on each individual’s dataset, in addition to hybrid algorithms trained on the group plus a proportion of each individual’s data, had improved performance but could only be constructed for a minority of participants. Implications This study used routinely collected data from a popular smartphone app to train and test a series of supervised machine learning algorithms to distinguish lapse from non-lapse events. Although a high-performing group-level algorithm was developed, it had variable performance when applied to new, unseen individuals. Individual-level and hybrid algorithms had somewhat greater performance but could not be constructed for all participants because of the lack of variability in the outcome measure. Triangulation of results with those from a prompted study design is recommended prior to intervention development, with real-world lapse prediction likely requiring a balance between unprompted and prompted app data.