NASA developed the Kepler Telescope using observations of transit events, which are smalldips in a star s brightness when a planet passes in front of it. Since transit events last only a fraction of the day, all stars must be monitored continuously, implying that their brightnesses must be measured at least once every few hours. The ability to continuously view monitored stars means that the field of view must never be blocked at any time of the year. Such conditions, especially in space, are difficult to predict, therefore, some cases of false positives and negatives occur when classifying exoplanets. In order to improve this classification, dealing with the problem of false positives and negatives, it was developed two types of predictive model approaches for exoplanet observations made by the Kepler telescope. For this, a Model Free system was used, based on machine learning models using Neural Networks and Neuro-Fuzzy Network, and also more mathematical models, such as Extremum Seeking Control. For the Neural Networks model, two of them were made, chosen and compared in Python: Perceptron and Multi-Layer Perceptron. Of these models, the one with the best results was chosen based on best peforming accuracy, precision, recall and RMSE metrics. For the Neuro-Fuzzy Networks model, it was developed in MATLAB two models with completely different databases. Of these models, the chosen one was with the best result involving accuracy metrics and RMSE. For the Extremum Seeking Control model, a system was made in MATLAB with the same noise as the data. This noise was then applied as a cost function. After that, a sinusoidal input was applied and with that, it was constructed an efficient control system to analyze the best response.