On exploring features engineering and machine learning algorithms in the domain of clinical-psychological

The primary goal of this research is to classify the treatment outcomes for patients through machine learning (ML) models. The study first detects multicollinearity issues using heatmap within the features in the data set and this entails a high correlation between the features which are detrimental to the models. In response to it, lasso (least absolute shrinkage and selection operator) and PCA (principal component analysis) are used to cope with multicollinearity and to perform the fine-tuning on feature selection and data dimensionality reduction. Whereas Lasso reduces the coefficients of insignificant features to zero, PCA reorients the dataset into principal components to achieve the maximum variances. After applying PCA and Lasso, the study utilized three ML techniques: SVM (support vector machine), logistic regression and decision tree algorithm to classify the treatment outcomes. Thus, from the analysis, the findings revealed that in terms of accuracy, precision, and recall, PCA improves the performance of non-linear models like SVM and Logistic Regression such that they outperform Lasso. It improves these models’ average ability to generalize. In case of Decision Tree, however, Lasso performed much better and benefited the algorithm in terms of accuracy and precision by providing the model with only necessary features.