Cluster Ensemble Method and Convolution Neural Network Model for Predicting Mental Illness

Abstract

One in every four individuals has a diagnosable mental disease in a year. Around 20% of children and adolescents have a mental health condition and often ignore it. It is found that 93% of youth use social media to communicate and engage, as it reflects their emotions, moods, and thoughts. As a result, machine learning algorithms may anticipate people's moods and emotions based on their postings and comments. On the other hand, psychometric tests use questions to determine how individuals think, feel, behave, and react. It is necessary to investigate a hybrid approach for identifying people's mental illness by combining social media inputs and psychometric tests, especially during a pandemic. The hybrid approach can combine the results from both the models to reflect on the user's digital & non-digital reactions to certain sensitive situations to determine their mental state. Hence, the present paper aims to develop a web framework that can forecast the emergence of mental illness in the future based on data from social media comments and real-time data from psychometric tests using machine learning algorithms. The proposed work includes the cluster ensemble method for social media posts and a convolution neural network model for psychometric tests. This model predicts mental illness with an accuracy of 87.05 percent. The individual can use this result to take the required precautions by visiting a psychologist.

Cluster Analysis of Japanese Whiskey Product Review Using K-Means Clustering

“…A cluster is defined by the position of the center in the n-dimensional space of the Attributes of the ExampleSet [21][22][23][24][25]. Some researchers use process clustering in their research to obtain valid results [26][27][28][29][30]. The similarity between Examples is based on a distance measure between them.…”

Machine Learning for Multimodal Mental Health Detection: A Systematic Review of Passive Sensing Approaches

“…Word count-related representations: Term Frequency-Inverse Document Frequency (TF-IDF), Bag of Words (BoW), n-grams, Term Frequency-Category Ratio (TF-CR) [379] Word2Vec embeddings, language models [115,116,118,124,128,130,140,143,144,148,185,186,188,198,217,374] Linguistic, Representations Readability metrics: Automated Readability Index (ARI), Simple Measure of Gobbledygook (SMOG), Coleman-Liau Index (CLI), Flesch reading ease, Gunning fog index, syllable count scores Textstat [380] [218,220] Linguistic…”

“…Lexicon-based representations [381] Depression domain lexicon [382], Chinese suicide dictionary [370] [120,135,189] Representations Sentiment scores, valence, arousal, and dominance (VAD) ratings NLTK [280], IBM Watson Tone Analyzer, Azure Text Analytics, Google NLP, NRC emotion lexicon [383], senti-py [384], Stanford NLP toolkit [281], Sentiment Analysis and Cognition Engine (SEANCE) [282], text SA API of Baidu Intelligent Cloud Platform [123], Valence Aware Dictionary and Sentiment Reasoner (VADER) [385], Chinese emotion lexicons DUTIR [386], Affective Norms for English Words ratings (ANEW) [283], EmoLex [? ], SenticNet [388], Lasswell [389], AFINN SA tool [390], LabMT [391], text2emotion [392], BERT [266] [ 20,54,61,86,110,115,118,119,121,123,[126][127][128]130,132,133,137,[143][144][145][146]148,[184][185][186]188,…”

“…ElasticNet regression [437], stochastic gradient descent (SGD) regression, Gaussian staircase model, partial least square (PLS) [438] regression (useful for collinear features), generalized linear models Learn relationship between features to predict continuous values (scores of assessment scales) or probabilities (correspond to output classes) [20,43,49,53,55,68,70,99,104,105,126,130,134,140,150,154,163,164,167,175,179,182,188,200,[211][212][213]219,222,223] SVM Find a hyperplane that best fits features (regression) or divides features into classes (classification), secondary model in score-level fusion [47,50,79,99,104,105,115,121,130,134,140,148,162,163,…”

“…Predict based on numerical input features [51,99,104,105,114,126,130,134,140,148,151,155,157,160,163,164,167,169,178,179,182,183,188,203,204,206,212,219,222,223] RF Secondary model that predicts from regression scores and binary outputs of individual modality predictions [71,81] Balanced RF [445] (RF on imbalanced data) Aggregate predictions of ensemble on balanced down-sampled data [209] XGBoost-based subject-specific hierarchical recall network Deduce subject-level labels based on whether the output probability of XGBoost at a specific layer exceeds a predetermined threshold [194] Stacked ensemble learning architecture Obtain the first level of predictions from KNN, naive Bayes, Lasso regression, ridge regression, and SVM, then use them as features of a second-layer logistic regression [123] Feature-stacking (a meta-learning approach) [303] Use logistic regression as an L1 learner to combine predictions of weak L0 learners on different feature sets [185] Table A9. Cont.…”