Among many serious global health issues in females, cervical cancer is ranking at fourth number. As seen in most of the cases, high persistent human papillomavirus (HPV) is the biggest reason of cervical cancer globally. Cancer is asymptotic in its earlier stages. Its screening rates are not very good, especially in the developing countries. This study proposes a comprehensive machine and deep learning based pipeline for predicting cervical cancer from patient’s medical history and other risk factors associated with cervical cancer. This approach integrates advanced data preprocessing using an autoencoder to impute missing values. It also dealt with the challenge of class biasness by using ADASYN oversampling technique. Principal component analysis is also implemented for dimensionality reduction while preserving most variance from the data. Three models are trained in deep learning, evaluated and compared to leverage the strengths of each implemented model. CNN and MLP are trained and tested individually as well as combined in the soft voting ensemble model to enhance the predicting accuracy of the final model. The findings demonstrated that the ensemble model secured the superior position with 98.45% accuracy, 100% recall, 96.71% precision, and 98.33% F1-score. Combining advanced preprocessing techniques with ensemble model delivered the most promising results highlighting effectiveness of the proposed pipeline. It offers a great decision-support approach to assist doctors in early prediction and avoiding unnecessary medical procedures for those who does not need them.