Many cognitive phenomena can be interpreted through the paradigms of complex and chaotic systems, allowing the use of mathematical models in data analysis. Some of the most important aspects of change processes in psychotherapy (e.g., discontinuous progress) can be explained in the context of their chaotic dynamics [1]. The phase transition of self-organizing systems (PT) is characterized by changes in various dynamic aspects of the client’s PT’s multivariate time series (e.g., a change in the mean or variance over time) [2]. I was interested in whether the use of machine learning (ML) models contributes to increased accuracy in detecting PTs. I investigated this by (a) analysing the specific options for introducing the ML model into the Pattern Transition Detection Algorithm (PTDA) [3], which contains several sub-algorithms for detecting PT, and (b) implementing the chosen option of introducing the ML model into PTDA and analysing whether the extension contributes to increasing the detection accuracy. I used two existing datasets, one containing a heterogeneous sample of 30 clients and the other of 40 participants. Both datasets consist of a time series of self-assessment questionnaires and diary entries. I used the first dataset to develop a model for detecting PT, and the second to demonstrate the feasibility of transferring the solution to other mental health time series data. For (a), I used the literature and analysis of the PTDA to list the options for implementing ML models and select the most appropriate one. For (b), I implemented a ML model and evaluated performance by comparing the accuracy of the expanded and original versions of the PTDA. The results showed that, compared to the original PTDA, the inclusion of a machine learning regression model contributed to a statistically significant increase in the R2 performance metric in detecting phase transitions. This points to the advantage of using machine learning models because of their ability to discover patterns and additional knowledge by learning from the data. Additionally, I proposed a systemic solution to guide the implementation of this system for utilization in psychotherapy practice, emphasizing the importance of close interaction between psychotherapists and computational methods. The proposed solution can serve as an effective tool in psychotherapeutic practice. With this work, I took an important step in the process of incorporating machine learning models to detect phase transitions in mental health.
|