The commercialization of perovskites for solar cells and LEDs requires considerable increase in their long-term stability. Environmental stressors, such as light, humidity, temperature, bias, and oxygen, cause their degradation, which restricts device lifetimes to far below those of state-of-the-art inorganic semiconductors. For effective commercialization, the large parameter space of perovskite compositions and synthesis conditions must be traversed to design optimal devices. Such devices must also undergo lengthy stability testing to prove their viability. Therefore, using a standard trial-and-error materials research approach halide perovskites becomes very time consuming. Machine learning (ML) is a promising tool to accelerate the influence of the aforementioned environmental stressors on perovskites’ luminescent properties. We acquire environmental photoluminescence (PL) measurements on Cs-FA perovskites under a variety of relative humidity conditions, and apply ML to uncover the effect of water of perovskite stability. To perform a time- series prediction, we implement recurrent neural networks (RNN), a class of neural networks with a historical state or “memory.” Briefly, the network forecasts transient PL responses using the environmental state (rH and temp) at a specific time point along with the historical state, which represents physical changes that have occurred within the film. Our network can be driven by weather forecast data, predicting future PL output with high time resolution (one point every 15 seconds) and providing insight into the optical response and stability of perovskite materials under various weather conditions. Our results showcase the potential of ML for halide perovskites expansion through the example of time-series prediction of optical properties.