This paper investigates the classification of customer feedback into emotions and examines document clustering to uncover themes within the feedback in Spanish using advanced NLP techniques , ML models, and Deep Learning embed-dings. We evaluate Logistic Regression, custom Neural Networks (cNN), and Support Vector Machine (SVM) models. Our experiments reveal that Logistic Regression and cNN achieve average accuracies of 97.13% and 98%, respectively, on the training set, and 85.17% and 86.12% on the test set. Various SVM kernels with two principal components (PCAs) resulted in training accuracies of 64.11%, 66.51%, 67.22%, and 46.29% for the linear , polynomial, RBF, and sigmoid kernels, respectively , and corresponding test accuracies of 62.20%, 62.20%, 64.11%, and 44.02%. A grid search identified the RBF kernel with C = 0.5 and γ = 10 as having the highest training and test accuracies of 66.51% and 64.59%. Using a 3-PCA SVM approach, training accuracies for the linear, polynomial, RBF, and sigmoid kernels were 66.63%, 66.39%, 70.93%, and 61.60%, and test accuracies were 62.68%, 63.64%, 70.33%, and * Equal contribution † 60.77%. The RBF kernel with C = 10 and γ = 1 achieved the highest training and test accuracies of 69.86% and 70.81%. These findings highlight the complexity of emotion classification in Spanish and demonstrate the potential for high accuracy using these models.