Application health is determined by calculating baseline deviation for one or more transactions based on past data having the same context as current data being examined. The performance data is generated in response to monitoring one or more applications that perform transactions during a time period. The performance data may include transaction metric data and the context data describing conditions under which one or more transactions were performed. A baseline is determined by predicting current or recent transaction performance data values using past transaction performance data associated with the same context. Based on the comparison of the actual and predicted data with the same or similar context, a deviation from the baseline value is determined for the transaction and reported, for example to a user through an interface.