A SURVEY OF CHILD MENTAL HEALTH DETECTION

ABSTRACT:

Child mental health is a critical area of concern that requires effective detection strategies to
ensure timely intervention and support. This survey aims to explore the current methods and
challenges in detecting mental health issues among children, with a focus on identifying gaps and
potential improvements in practice. Proactive detection of mental problems is a critical step in
addressing this issue. However, accurate detections are hard to achieve due to the inherent complexity
and heterogeneity of unstructured multi-modal data generated by campus life. Machine Learning and
AI: These technologies are used to analyze behavioral data and predict mental health conditions. Various
machine learning models, such as neural networks and support vector machines, are employed to identify
patterns associated with mental health disorders.Multimodal Sensing: This approach integrates data from
multiple sources, such as wearable devices, mobile apps, and social media, to provide a comprehensive
assessment of a child's mental state.

INTRODUCTION:

Depression in children is a mental health condition that affects the mind, mood and
development of the children. If depression persists without treatment for a long time, children can fall
behind in their usual activities such as learning[1]. MENTAL health can be defined as “a state of
wellbeing in which individuals realize their abilities to cope with the normal stresses of life, work
productively and fruitfully, and contribute to the community”. Mental health monitoring is a potent
tool in supporting the treatment of mental issues. Several sensing methods have been reported in the
literature for mental health monitoring. These methods can be invasive or noninvasive (contactless)
based on their interaction with the subject. Invasive methods are more reliable and efficient than
noninvasive ones. Traditionally, invasive methods were being used to monitor the mental state via
electroencephalogram (EEG) features in hospitals[2]. A framework, named CASTLE (eduCational
dAta fuSion for menTaL hEalth detection), is proposed to achieve an accurate and effective detection,
which includes three parts. First, representation learning is used here for the effective fusion of
students’ multi-modal information, including multi-view social network embedding, physical
appearance representation, and academic performance representation. In this part, we propose MOON
(Multi-view SOcial NetwOrk EmbeddiNg), a multi-view social network embedding algorithm, to
embed students’ heterogeneous social relations effectively. Moreover, a convolutional neural network
(CNN)-based auto-encoder is used to embed students’ ID photos to obtain an accurate representation
of their physical appearance. In addition, we introduce the method of combining the variant of one-
hot encoding and autoencoder to overcome the heterogeneity of students’ academic performance.
Second, a synthetic minority oversampling technique (SMOTE) algorithm is used to mitigate the
effects of label imbalance. Finally, a deep neural network (DNN) model is used for the final
detection.contributions can be summarized as follows:

• Detection framework CASTLE for students’ mental health detection through fusing multi-modal
information generated from campus life.

• Multi-view social network embedding algorithm MOON, eliminating the information redundancy
across different views through simplifying the strategy of federated embedding.
 • We conduct comprehensive experiments on a real-world educational dataset, and the extensive
results demonstrate the promising performance of the proposed methods in comparison with an
extensive range of state-ofthe-art baselines.

Machine learning techniques implemented in child mental health detection surveys often include
Supervised Learning Algorithms like decision trees, random forests, and support vector machines are trained
on labeled datasets to classify mental health conditions based on input features such as behavior, speech
patterns, and physiological data.Unsupervised Learning: Clustering methods (e.g., k-means, hierarchical
clustering) group children with similar behavioral patterns, aiding in identifying at-risk groups.Deep Learning:
Neural networks, especially Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs),
process complex data such as video, audio, and sequential data to detect subtle indicators of mental health
issues Natural Language Processing (NLP): Techniques like sentiment analysis and topic modeling analyze
text from social media, surveys, and conversations to identify signs of anxiety, depression, or other mental
health concerns.Reinforcement Learning: Used in adaptive interventions where the system learns to provide
personalized recommendations or therapeutic activities based on the child’s responses and progress.

Machine Learning Algorithms for Depression

Multimodal techniques in child mental health detection surveys combine data from various sources to provide a
holistic view of a child's mental state. Key approaches include Wearable Devices: Collect physiological data (e.g.,
heart rate, sleep patterns) to monitor stress and anxiety levels.Mobile Apps: Track behavior and mood through self-
reports and activity logs.Audio and Video Analysis: Use machine learning to analyze speech patterns, facial
expressions, and body language for signs of mental distress.Social Media Monitoring: Analyze language and
interaction patterns using natural language processing to detect changes indicative of mental health issues.[4]

ILLUSTRATION OF RESEARCH

Machine Learning for Multimodal Child Mental Health Detection

LITERATURE SURVEY:

There are researches that are done on child mental health detection

1. Umme Marzia Haque,Enamul Kabir,Rasheda khanam “recent advances in contactless

sensing technologies for mental health monitoring” published in 2021 in this study
proving accurate and efficient classifier to detect depression using YMM, a large
dimension dataset of mental heatlh student.
2. Muhammad Nouman,Sui Yang khoo, M.A. Paarvez Mahmud , Abbas
Z.Kouzan “Recent Advances in Contactless Sensing Technologies for
Mental Health Monitoring” published in January 2022 this article
presented a detailed and uptodate review of the contactless sensing
technology and their mental health monitoring application.
3. Ziyun Jiang,Xuejia Ma ”research on mental health of migrant children based on
wireless communication detection technology” published in 2022 based on the
important life and emotional experiences of migrant children determine that they will
face more psychological development problems than ordinary urban children.
4. Teng guo,Wnho Zhoa,Amr Tolba,Feng Xia ”multimodal educational data fusion for
students’ mental health detection” published in july 2022 related to investigate an
important problem of the detection of student’s mental health.
5. Christian nash, Rajesh nair, syed Mohsen naqvi ”machine learning in adhd and
depression mental health diagnosis: a survey”published on 2023 related to the survey
to detail about machine learning applications in mental health detection
6. Bryn c.loftness, Julia Halvorson-Phelan , Aisling O’Leary and others“the champ app:
a scalable mhealth technology for detecting digital phenotypes of early childhood
mental health” published in 2024 related to e provide an accessible, digital tool to
collect physiological and behavioral data during brief, structured mood induction
tasks and an open-source platform to engineer objective features related to childhood
emotional and behavioral health.

METHODOLOGY:

1. ML algorithms.[1]
o Boruta algorithm. Boruta is a wrapper algorithm built around the RF classifier to
determine the relevance and importance of features in relation to the target
variable . The aim of this algorithm is to remove unusable features in relation to
the target variable which can create unnecessary noise.
o RF. An RF algorithm has hyper-parameters indicating the number of trees and
maximum tree depth considering how many interactions in the model are
evaluated and the decisionmaking rules are the parameters. It constructs
numerous decision trees and combines them to get a more accurate and reliable
prediction.
o TPOTclassifier. The TPOTclassifier provides intelligent machine-learning
searches which include supervised classification models, preprocessors, selection
strategies and any other science-learn API-assessment estimators or transformers.
It can find the best parameters and model ensembles using genetic programming.
 o XGB. XGB uses gradient boosting in the random forest. The gradient boosting is
achieved by giving rankings for each tree leaf and creating new trees. Depending
on the performance of the previously created trees, it assigns different weights for
each tree. This contrasts with traditional approaches for boosting.

Functional pattern of the proposed method for depression detection

Replace network embedding with[4]

Deepwalk : Deepwalk is a classic embedding algorithm that is widely used in the

field of network representation.

MANE : MANE is a multi-view network algorithm, which inspired us to propose the

MOON algorithm.

Replace final detection model with current popular algorithms shown as follows:

• Support Vector Machine (SVM) : SVM is a classic algorithm and is widely used
in the field of data mining.

• Random Forest (RF) : is a classic ensemble algorithm that achieves good

performance in various applications.

• XGBoost : XGBoost is a boosting-tree-based method and is widely used in various

data mining scenarios with good performance.

SMOTE algorithm to augment data in order to improve generalization performance.

DATASET:

DISCUSSION

Despite advancements in child mental health detection, significant gaps remain. These gaps
can hinder early diagnosis and timely intervention, which are crucial for addressing mental
health issues in children effectively. This discussion highlights the major gaps in current
detection methods, including traditional approaches, technological tools, and their
integration.
CONCLUSION

Early detection of mental health issues in children is vital for timely intervention and better
outcomes. While traditional methods remain essential, the integration of advanced
technological tools offers significant potential to enhance the accuracy, accessibility, and
effectiveness of mental health assessments. Addressing the challenges of accuracy, privacy,
ethical use, and integration with healthcare systems will be crucial for the successful
implementation of these technologies. The future of child mental health detection is
promising, with the potential to revolutionize how mental health issues are identified and
managed, ultimately contributing to the well-being of children globally.

By synthesizing traditional and new approaches, continuing interdisciplinary research, and

ensuring ethical implementation, we can improve the early detection and treatment of mental
health issues in children, paving the way for healthier futures.