Predicting Symptoms of Amotivation and Anhedonia among University Students with a Novel Oversampling Method

Dang Nguyen; Bao Duong; Arun Kumar; Dat Phan-Trong; Julian Berk; Taylor Braund; Kien Do; Debopriyo Bal

cs.LGAug 31, 2026

Predicting Symptoms of Amotivation and Anhedonia among University Students with a Novel Oversampling Method

Authors: Dang Nguyen, Bao Duong, Arun Kumar, Dat Phan-Trong, Julian Berk, Taylor Braund, Kien Do, Debopriyo Bal,

Organizations: Applied Artificial Intelligence Initiative (A2I2), Deakin University, Geelong, VIC, Australia · Black Dog Institute, University of New South Wales, Sydney, NSW, Australia

Abstract

University students experience disproportionately high rates of common mental health conditions, such as depression, which can impair learning, social functioning, and overall well-being. Within this context, symptoms of amotivation (i.e. loss of motivational drive) and anhedonia (i.e. diminished interest or pleasure) are particularly debilitating, yet they frequently go undetected. Developing new approaches to identify students with prominent amotivation and anhedonia could enable earlier and more targeted intervention. Machine learning (ML) methods have increasingly been used to classify individuals according to symptom severity. However, these ML models often suffer from class imbalance, where the majority of cases fall in the low-symptom group and relatively few in the high-symptom group. This imbalance can reduce model accuracy and bias predictions. To address this, studies commonly employ the popular oversampling strategy SMOTE. However, SMOTE has a notable limitation: it may generate invalid values for nominal variables. In this paper, we introduce a novel and effective oversampling method that addresses this shortcoming. Our approach leverages a predictive model to generate nominal variables, rather than interpolating them. We validate our method on a large-scale GPS location dataset collected from university students and demonstrate that it is significantly better than existing oversampling approaches in predicting elevated symptoms of amotivation and anhedonia.

Figures & tables

Figure 1: Training and evaluation phases of an oversampling method. Training: the oversampling method learns from the imbalanced dataset

{\cal D}_{train}

to generate synthetic minority samples

\hat{{\cal D}}_{minor}

to construct the rebalanced dataset

\hat{{\cal D}}_{train}

. Evaluation:

\hat{{\cal D}}_{train}

is used to train ML classifiers, and the classifiers are evaluated on a held-out test set

{\cal D}_{test}

to compute performance metrics (e.g. AUC-scores). A better score implies a better oversampling method .

Figure 2: Our SMOTE-PRED has three steps. First, we apply SMOTE to the imbalanced dataset

{\cal D}_{train}

to generate synthetic minority samples

\hat{{\cal D}}_{minor}

. As SMOTE treats all variables as continuous variables, it generates continuous values for the nominal variable Treatment_ID . Second, from

{\cal D}_{train}

, we extract continuous and nominal variables, and use them as features and labels respectively to train a predictive model. Finally, we extract continuous variables from

\hat{{\cal D}}_{minor}

, and use them as input features for our trained predictive model to predict values for the nominal variable Treatment_ID . We then replace generated continuous values in Treatment_ID by SMOTE with our predicted integers (i.e nominal values).

Table 1: Some fake examples of our GPS location dataset for an illustrative purpose. Table (a) shows two students along with their visited locations. Each student has a unique ID and multiple GPS locations (in terms of latitude and longitude). Each of them is associated with a timestamp. “Horizontal Accuracy” and“Vertical Accuracy” variables measure the accuracy range (in meters) of the recorded locations. Table (b) shows the amotivation scores of these two students. Each score

s

is in a range of

[0,9]

, where

s<5

is “ no or minimal symptoms of amotivation ” whereas

s\geq 5

is a “ high symptoms of amotivation ”.

Panel of Figure 3 — Figure 3: Distributions of amotivation scores (a) and labels (b). Among 482 students, 80% of them are “low amotivation” (i.e. majority class) and 20% are “high amotivation” (i.e. minority class).

Student	Date	$n_{location}$	$dist_{travel}$
abcxyz12	15/11/2021	2	4.01
abcxyz12	16/11/2021	1	0
12abc456	19/11/2021	1	0

Table 2: Our fake examples in Table 1 (a) are represented with two features “the number of visited locations”

n_{location}

and “the average distance traveled”

dist_{travel}

Panel of Figure 4 — Figure 4: For an illustrative purpose, we show fake mobility patterns of a student with “low amotivation” (a) and a student with “high amotivation” (b). The x-axis shows the date while the y-axis shows the number of visited locations in one day.

Figure 5: Distributions of variables

n_{location}

and

dist_{travel}

in each trial. The y-axis shows the mean value and its standard error.

Student	$n_{location}$			$dist_{travel}$			trial_id
	day1	…	day15	day1	…	day15
abcxyz12	3.0	…	3.3	2.8	…	1.3	1
12abc456	2.0	…	3.0	8.0	…	4.3	3

Table 3: The final tabular format of our GPS location dataset has 30 continuous variables and one nominal variable trial_id .

Classifier	Hyper-parameters
kNN	n_neighbors : {1, 2, …, 10}
SVM	kernel : {linear, rbf} gamma : {0.1, 0.5, 1.0} C : {0.1, 0.5, 1.0, 10.0}
DT	max_depth : {1, 2, …, 12}
RF	n_estimators : {1, 10, 50, 100} max_depth : {1, 2, …, 12} max_features : {sqrt, log2, None}
XGB	n_estimators : {1, 10, 50, 100} max_depth : {1, 2, …, 12} max_features : {sqrt, log2, None}

Table 4: Hyper-parameters of ML classifiers to predict student amotivation.

AUC-score	Imbalance	AdaSyn	SMOTE	SMOTE-NC	CTGAN	TVAE	SMOTE-PRED
kNN	0.5221	0.6065	0.6346	0.6498	0.5382	0.5436	0.6735
kNN	(0.011)	(0.035)	(0.022)	(0.028)	(0.015)	(0.013)	(0.025)
SVM	0.4949	0.5854	0.5817	0.5572	0.4806	0.5963	0.6198
	(0.003)	(0.025)	(0.015)	(0.019)	(0.009)	(0.019)	(0.024)
DT	0.5310	0.5372	0.4992	0.5237	0.5167	0.5012	0.5491
	(0.019)	(0.014)	(0.024)	(0.027)	(0.013)	(0.016)	(0.023)

Table 5: AUC

\pm

(std) of each oversampling method combined with five popular ML classifiers. Bold and underline indicate the best and second-best methods.

Figure 6: AUC-scores of our method SMOTE-PRED and other SMOTE variants.

Figure 7: AUC-score vs. the number of imputed days.

Imputation method	AUC
linear	0.6735 (0.0246)
nearest	0.5976 (0.0259)
nearest-up	0.5938 (0.0254)
zero	0.5929 (0.0215)
s-linear	0.6735 (0.0246)
previous	0.5929 (0.0215)

Table 6: AUC-score vs. imputation methods.

Features	AUC
{ $n_{location}$ , $dist_{travel}$ , trial_id }	0.6735 (0.0246)
– $n_{location}$	0.5836 (0.0203)
– $dist_{travel}$	0.4697 (0.0275)
– trial_id	0.6360 (0.0155)

Table 7: AUC-score vs. features. “–

n_{location}

” means that we remove all 15 features related to

n_{location}

(see Table 3 ).

Figure 8: Feature importance. A higher score indicates a more important feature.

Class-Imbalanced Learning

Class-imbalanced learning develops classifiers that recognize rare classes without sacrificing performance on common ones.

See all 260 papers

SMOTE-VAR: An Uncertainty-Aware Oversampling Method for Predicting Depression Remission in University Students
Aug 31, 2026Dang Nguyen, Arun Kumar A, Taylor A. Braund +8Class-Imbalanced Learning Mental Health
Applied Artificial Intelligence Initiative (A2I2), Deakin University, Geelong, VIC, Australia · Black Dog Institute, University of New South Wales, Sydney, NSW, Australia
OverNaN: NaN-Aware Oversampling for Imbalanced Learning with Meaningful Missingness
May 12, 2026Amanda S BarnardClass-Imbalanced Learning Synthetic Data Augmentation
School of Computing Australian National University Acton, ACT 2601
Beyond Imbalance Ratio: Data Characteristics as Critical Moderators of Oversampling Method Selection
Apr 6, 2026Yuwen Jiang, Songyun YeClass-Imbalanced Learning Classification
School of Artificial Intelligence, Guangzhou Institute of Science and Technology, Guangzhou, China

Aug 31, 2026J/K move · Enter open · S save

SMOTE-VAR: An Uncertainty-Aware Oversampling Method for Predicting Depression Remission in University Students

Dang Nguyen, Arun Kumar A, Taylor A. Braund, Wu Yi Zheng+7

Applied Artificial Intelligence Initiative (A2I2), Deakin University, Geelong, VIC, Australia · Black Dog Institute, University of New South Wales, Sydney, NSW, Australia

Class-Imbalanced Learning Mental Health Clinical Outcome Prediction

Open paper AskPDF arXiv

University students experience disproportionately high rates of common mental health conditions, such as depression, which can impair learning, social functioning, and overall well-being. Although lifestyle interventions such as mindfulness and physical activity can reduce the symptoms, many do not achieve symptomatic remission. Developing new approaches to identify students with poor outcomes could enable earlier and more targeted intervention. Machine learning (ML) methods have increasingly been used to predict remission in depressive patients. However, these ML models often suffer from class imbalance, where there may be an unequal proportion of people in the remitted group relative to the non-remitted group. This imbalance can reduce model accuracy and bias predictions. To address this, studies commonly employ the popular oversampling strategy SMOTE. However, SMOTE has a notable limitation: it may generate invalid synthetic minority samples. In a clinical context, these false positives can lead to incorrect risk stratification, potentially delaying necessary escalated care for patients unlikely to remit. In this paper, we introduce a novel and effective oversampling method that addresses this shortcoming. Our approach leverages the variance function of a Gaussian process to estimate the uncertainty of generated minority samples to reduce false positives. We validate our method on a depression dataset collected from university students and demonstrate that it is better than existing oversampling approaches in predicting remission (i.e., treatment outcome). By improving the reliable identification of non-responders, our method provides a robust computational tool to help clinicians rapidly pivot to adjunctive therapies, thereby personalizing and optimizing mental health care pathways.

Abstract

Class-Imbalanced Learning

Explore similar work

SMOTE-VAR: An Uncertainty-Aware Oversampling Method for Predicting Depression Remission in University Students

OverNaN: NaN-Aware Oversampling for Imbalanced Learning with Meaningful Missingness

Beyond Imbalance Ratio: Data Characteristics as Critical Moderators of Oversampling Method Selection