This study aims to compare the performance of computational intelligence methods to classify 3D vectorcardiographic (3D-VCG) data in normal (NORM), myocardial infarction (MI), and ST-T change (STTC) groups. 3D-VCG is advantageous over conventional electrocardiography (ECG), particularly in detecting patterns unavailable in single-lead. Unlike standard ECG, with unidimensional leads, 3D-VCG captures spatial information, allowing a more comprehensive analysis of cardiac depolarization. The 12 ECG leads from the PTB-XL Database were converted into XYZ leads of 3D-VCG using the Kors matrix. Data from 7,146 patients were randomly selected to produce a balanced dataset across NORM, MI, and STTC groups (2,382 each). Three machine learning models were trained with 70% of the data and tested with the remaining 30%: the multi-layer perceptron (MLP), the convolutional neural network (CNN), and long short-term memory (LSTM). The input data was the 3D-VCG with 350 samples/lead. The MLP model achieved 99.99% accuracy during the training and 99.80% in testing. The CNN model achieved 96.07% accuracy during the training and 83.26% in testing. The LSTM model achieved 99.90% accuracy during the training and 89.00% during the test. Applying the 10-fold cross-validation approach only for the LSTM model allowed us to obtain an average accuracy of 94.0 ± 1.8%. Although this approach uses 90% of data for training in each fold, LSTM does not reach the performance of the classical MLP network. In conclusion, this study provides an effective framework for the automated detection of MI and STTC on ECG.