The study aimed to address the global challenge of cancer-related fatalities by investigating the feasibility of identifying or predicting the early-stage presence of three distinct forms of cancers, colon, thyroid and urothelial carcinoma, via the analysis of raw DNA sequences. The data, sourced from the NCBI database, underwent a series of pre-processing techniques, including kmer analysis, under-sampling and count vectorization. Subsequently, machine learning algorithms, including logistic regression and multinomial Naive Bayes, were implemented on the pre-processed data with logistic regression demonstrating superior accuracy of 80.10% with calibration and 78.54% without calibration. To enhance the model's extrapolative capabilities, the logistic regression model was further calibrated utilizing the sigmoid method. The final model was deployed through the utilization of the open-source streamlit package.