Artificial Intelligence, Philosophy of science, mental health, adaptive technology, induction, research methods
Statistics reveals exceptional incidence of mental health disorders around the world, in Europe [2, 3] and in Norway . Dealing with such mental health patients can be economically [5, 6], physically and emotionally challenging. Several types of researches have been carried out to facilitate automated and personalized screening and treatment to such mental health patients. Human Brain Project , eMEN , and INTROMAT  are among the European projects that address such mental health problems. In particular, the goal of INTROMAT is to offer personalized treatment for patients suffering from mental and neurological disorders. My project, entitled as Personalized Treatment of Mental Health Problem using Voice is a part of INTROMAT project, where I intend to use human voice as biological marker along with Artificial Intelligence algorithms to detect if a person is suffering from any mental health morbidities. This automation of detecting mental health illness helps to continuously monitor the patients condition on one hand, and create personalized Internet Based Treatments (IBT) on the other.
To relate the main topics of this project, namely Personalized Treatment of mental health problem using adaptive technology, to philosophy of science, a short overview of philosophy of of science on the one hand and building adaptive system for mental health problem on the other hand is required. The following two sections are devoted to give a respective overview. Section 2 looks at background of philosophy of science and Machine Learning. Section 3 outlines the main research questions of the project. Section 4 outlines scientific methodology. Section 5 performs the classification of this Ph.D. project based on Philosophy of science.
The recent rapid growth in Artificial Intelligence (AI) has opened up unprecedented possibilities in analyzing and predicting diverse phenomena including humanities, social and cognitive sciences, finances, healthcare, robotics and other areas of natural sciences. AI is a broader concept that consists of Deep Learning (DL) and Machine Learning (ML). DL is a subset of ML, which is also a subset of AI.
ML is is an important aspects of data science where the goal is to make a rule which takes a data Set A and predicts a data set B. For example, if we have set of voice of patients who are depressed, we can try to predict a new patient has depression or not. The idea is that if we take a series of complicated functions, then we can find a function which makes accurate predictions. The notable aspects of ML is the way models are trained and evaluated. A data set is divided into training data set and test data set. By design, the models are always constructed using training data set. And the model is evaluated using test data set. The goal is to create an optimal model that can predict results for the testing set when it has only seen the training set. This task is both extrapolation and interpolation. A good ML model should be able to understand the pattern in the training data and should be able to recognize the same pattern in new data.
One drawback of ML methods is that even an accurate model does not provide any additional understanding of the problem. At best, we can find there is relation between X and Y but the model does not explain what is the relationship between them. This understanding is a central matter of philosophy of science. The classic theory of Kuhn  says, science is the process of observing the world, making hypothesis based on those observations, and evaluation of those hypothesis by making new measurements. However, the theory does not take uncertainty into account. All the measurements have certain degree of uncertainty which makes it impossible to sate anything by certainty. ML only cares about accuracy and not about certainty. By these principles, we can proclaim that science is about making accurate predictions as stated by the Bayesian perspective to philosophy of science, we try to find the most likely solutions to scientific problems.
The primary aim of this research is to come up with adaptive technology that can be used in both clinical setup and home-based setup for personalized assessment and treatment of mental health problems. Now we present our research questions and in the following sections we will present how we address them.
RQ1 What kind of voice signal features are taken into considerations in the detection of mental health problems? How can these features be categorized?
RQ2 What are the approaches and methodologies used in data collection from the mental health patients? How can these methods be applied to clinical diagnosis?
RQ3 What type of machine learning algorithms that has been applied on physiological dataset for detecting mental health problems such as depression, social anxiety, bipolar disorder, ADHD and others in the state of art works? What evaluation metrics has been used to measure the performance of these algorithms?
RQ4 How can we utilize MDSE techniques to build an adaptive system that uses voice signal to diagnose mental health problems and provide personalized treatments? Can the framework be utilized both in a clinical setup and home-based treatment?
An overview of overall organization of the project is illustrated in Figure 1. The overall phase of the project is divided into five phases.
Literature review shows that authors have considered parameters including video, text and physiological sensors to detect the stress level. In order to get the detailed overview of state-of-art researches and background knowledge on the problem domain, a systematic survey will be done. This will provide the groundwork for answering RQ1 and publishing research paper RP1. This initial step of the project is aimed at formulating a set of observations to understand problems and the state-of-art methods for solving them. The comprehension achieved in this step facilitates formulating hypotheses for our research questions.
The literature review provides better insights into the existing problems and solutions, identifies new opportunities and comes up with new ideas. These observations forms the basis for developing hypotheses. Based on these hypotheses we will have predictions that we expect to acquire in our experiments.
This step provides experimental design to test and evaluate our hypotheses. The end users can create a test session using mobile app which stores the sensor data in their local phone memory and can be synced to the server with a proper consent to respect their privacy. One of the most important aspects of the research is data collection. Preliminary research showed that there is a lack of training data that can be used for training a machine learning model. It is one of the important research questions in RQ2, regarding how data can be collected from mental health patients in a clinical setup. The work will provide the basis for answering research question RQ2 and grounds for publishing research paper RP2.
As explained in the Scientific Methodology, the main task is the create a ML model that can detect mental illness in patients. Basically, we start by proposing hypothesis which is the initial step in Popper rationalism. We use various parameter estimation and uncertainty quantification for different algorithms to conclude if a given hypothesis is true or false. Based on the nature of my Ph.D. project, I claim to classify the project in the following categories.
As mentioned in the Section 4, this project involves collection of observed data and use these data to create model that can induce if a new data follows similar pattern as the observed data. Hence, I claim the Ph.D. project follows inductive approach.
In the INTROMAT project, we research about five cases including ADHD, Depression, Bipolar Disorder, Social Anxiety and gynecological cancer. Hence, the research strategies clearly includes case study. Moreover, based on the detection of mental health conditions of the patients, we will be doing form based surveys and experiments.
I claim the Ph.D. project involves mixed method of research choices including both qualitative and quantitative research. To back up the statement, it is already said that we will be collecting voice data from users. Once the voice data is sampled and processed, it is numerical data. Moreover, we will be doing surveys, interviews with the patients, that qualifies for qualitative data. In addition to this, the prediction done by the ML model will be numerical value ( for example depression level ).
Saunders et al. (2007)  says, time horizons are needed for the research design and independent of the research methodology used. This project follows cross sectional as the research is carried to a specific time frame and cross sectional time horizon is used. This is due to the fact that, mental health patients requires continuous monitoring over a period of time and expected to get better and live a normal life.
In this paper, I outlined the philosophical aspects of my Ph.D. project with respect to research methodologies, research approaches, research strategies, and research choices.
This essay is a part of the Ph.D. mandatory course PCS902- Philosophy of Science and Research Methodologies. The opinions, findings, discussions, recommendations, and conclusions illustrated in this chapter are those of the authors and do not reflect the views of any agencies.