The use of machine learning for disease diagnosis is gaining popularity due to its ability to process data and provide accurate results; however, its optimization remains a challenge. In the case of Chagas disease, endemic in Latin America and which has emerged as a health problem in more urban areas, early and accurate diagnosis is essential to prevent cardiac complications, since an estimated 65 million people are at risk of contracting it. This study used a database of 292 subjects distributed into three groups: healthy volunteers (Control group), asymptomatic Chagasic patients (CH1 group) and seropositive Chagasic patients with incipient heart disease (CH2 group). A densely connected neural network was used to classify them into the group to which they belonged. The network received as input the Approximate Entropy values of each individual, which were calculated from the 24-hour circadian profiles every 5 minutes (288 RR subsegments). In addition, time series data augmentation algorithms were applied during the training phase to improve the classification results. This approach allowed to reach 100% accuracy and precision, validated by the ROC curve with AUC values of 1. Thus, the efficiency demonstrated by the neural network suggests that increasing the amount of training data may be crucial to optimize the early diagnosis of cardiac involvement that may develop Chagas disease, and consequently, it could be a determining factor in refining machine learning in this area.