Joshua Curtiss, an assistant professor of applied psychology at Northeastern University, is pioneering the use of machine learning to forecast human emotions. His research explores whether past emotional data can accurately predict a person's future mood, aiming to revolutionize mental health care by moving from generic treatments to personalized interventions. In a pilot study involving 34 participants with diagnosed emotional disorders, researchers collected self-reported data on four specific emotions: contentedness, cheerfulness, sadness, and anxiousness. Participants provided feedback five times daily over a two-week period using a seven-point scale. This data was analyzed by six different machine-learning models, ranging from simple averaging techniques to complex neural networks that mimic brain processing. The results indicated that individual models generally outperformed group-level benchmarks in predicting emotions up to one day in advance. Notably, the most accurate model varied by emotion. Models relying on past performance trends worked best for predicting contentedness and cheerfulness, while ensemble models, which combine the output of multiple individual models to create a composite prediction, were more effective for sadness and anxiety. Curtiss emphasizes that emotional disorders like depression and anxiety manifest uniquely in every individual. Current treatment approaches often rely on a one-size-fits-all strategy that may not address the specific factors driving a patient's distress. By identifying patterns specific to a single person, machine learning could enable providers to offer proactive, tailored support. This might involve alerting individuals to potential emotional shifts days or weeks in advance, giving them time to prepare or adopt coping habits to mitigate negative outcomes. The study remains in the proof-of-concept stage. Curtiss acknowledges significant challenges, including the unpredictability of external life events such as job changes or health news, which can drastically alter a person's mental state. He compares emotional forecasting to weather prediction: it is inherently probabilistic and subject to the butterfly effect, where small initial differences can lead to large, unpredictable changes over time. Consequently, the goal is not to create a perfect model capable of predicting a person's mood years in advance, but rather to generate useful, short-term forecasts one to two weeks out. Curtiss stresses the need for responsible implementation, recognizing that even the best models will occasionally be wrong. Don Robinaugh, an associate professor in the same department who was not involved in the study, praised the research for embracing human complexity. He noted that while the work faces hurdles, its potential to improve care is enormous. The research team is now planning to expand the study to include larger, more diverse populations, longer timeframes, and additional data sources such as smartphone and wearable device metrics, provided participants' subjective reports are supplemented with objective behavioral data. This expansion aims to refine the accuracy of these forecasting tools and bring the vision of personalized mental health support closer to reality.