The COVID-19 pandemic, starting in Wuhan, China in December 2019, led to widespread health and economic challenges, causing millions of deaths globally. Beyond physical health, it triggered a mental health crisis, especially during lockdowns. To understand and address this, a study collected data using 90 features during the lockdown period. Machine learning (ML) was employed to detect key features impacting mental health crises. Three ML algorithms—random forest, random tree, and multilayer perceptron—were chosen. Random forest, known for robustness, achieved 97.58% accuracy. Random tree, a supervised algorithm with decision trees, yielded 93.24% accuracy. Multilayer perceptron (MLP), an artificial neural network, achieved 94.20% accuracy by learning nonlinear relationships. A 10-fold cross-validation method was used to evaluate these ML models, enhancing performance by reducing bias and overfitting. It involves dividing data into ten subsets, training on nine, and evaluating the remaining, repeating this ten times to estimate true performance on unseen data.
Top1. Introduction
The first known case of COVID-19 occurred in December of 2019 in Wuhan, China. Still, the exact origin of COVID-19 is a matter of investigation. Scientists believe the virus originated in bats and then spread to humans through an intermediate animal host in Wuhan's Huanan Seafood Wholesale Market. It is not clear whether the virus originated from that market or spread through an infected person. On February 11, 2020, the WHO (World Health Organization) named the virus SARS-CoV-2, and on March 11, 2020, the WHO declared it a worldwide pandemic. COVID-19 hugely impacted the world, and due to this, millions of deaths occurred, which devastated the world economy. As per the latest report of November 26, 2023, WHO confirmed 6,981,263 deaths globally?
COVID-19 has not only created a health crisis but also caused a mental health crisis among people. A mental health crisis is an event when someone experiences severe emotional distress in everyday life. Especially during the lockdown period, people experienced heavy mental stress, which caused mental health crises. During this period of lockdown, we collected data from people. We gave them 90 features to fill, and they filled those as per their problem. Conventional procedures sometimes fail to detect which features are important for mental health crises. To detect important features that deeply impact mental health, we use machine learning algorithms (Khang & Medicine, 2023).
The choice of machine learning algorithms plays an important role in finding the exact features that are responsible for mental health crises. Here we selected three important algorithms Random Forest, Random Tree, and Multilayer Perceptron were used to detect those features that gave us the most accuracy. Random forest belongs to the ensemble learning technique. It is a versatile and widely used machine-learning algorithm that is known for its robustness and accuracy. It can handle both classification and regression tasks. Random Forest works on three core principles: bagging (bootstrap aggregation), random subspace sampling, and majority voting. Using Random Forest, we get almost 97.5845% accuracy.
Second, the random tree algorithm is a supervised machine learning algorithm. It works by constructing a multitude of decision trees at training time. It gives an output of the mode of predictions of the individual trees. This algorithm also works with Bootstrap aggregating (bagging), feature randomness, and the tree pruning method. The random tree gave us 93.2367% accuracy. Third, we choose a multilayer perceptron algorithm (MLP) that is a supervised machine-learning algorithm. It is used to learn nonlinear relationships between input and output data (Khang & Ragimova et al., 2024).
MLP is a feedforward artificial neural network (ANN) that consists of multiple layers of neurons. It is trained with a backpropagation algorithm, which is a gradient-based optimization algorithm. It iteratively adjusts the weights of the connections between the neurons in the network. MLP consists of three types of layers 1) the input layer; 2) the hidden layer; and 3) the output layer. Neurons in MLP use some activation functions like the sigmoid function, hyperbolic tangent function, and rectified linear unit (ReLU) to introduce nonlinearity into the network. Through the multilayer perceptron algorithm, we got 94.2029% accuracy.
Here, we used the resampling method of 10-fold cross-validation that is used to evaluate the performance of a machine-learning model. It usually divides the training data into 10 folds or 10 subsets of approximately equal size. Then, we train the model on the nine folds, and the evaluation is done on the remaining fold. We repeat this process 10 times, where each fold is used as the test set once. The average performance of the model across all 10 folds is then used as an estimate of its true performance on unseen data.10-fold cross-validation is an important tool in machine learning to reduce bias and overfitting. When a machine learning model is trained on data that is not representative of the data it will be used on in the real world, bias occurs. Overfitting occurs when a machine learning model learns the training data too well and is unable to generalize to new data.10-fold cross-validation helps to reduce bias by using a different test set each time the model is trained. It helps to reduce overfitting by making the model less likely to memorize the training data.