Covalent binding of chemicals to skin proteins represents the Molecular Initiating Event (MIE) of the skin sensitization process. We attempted to construct in silico models for predicting the reactivities of chemicals to cysteine measured by the Direct Peptide Reactivity Assay (DPRA) as a screening tool for skin sensitization potential of chemicals since there was no readily available in silico prediction model for reactivity classes of the DPRA. We used a dataset of 211 chemicals compiled based on the chemical reactivity to cysteine in the DPRA for model construction, and each chemical was classified as “Minimal-Low” or “Moderate-High” reactivity according to the percent cysteine depletion value in the DPRA. We constructed two independent classification models using two machine learning algorithms named Random Forest (RF) and Graph Convolutional Network (GCN), and a consensus model adopting prediction results when both of the GCN-based and the RF-based models were matched was also constructed. Performance evaluation showed that the RF-based model showed higher specificity than the GCN-based model and the GCN-based model showed higher sensitivity than the RF-based model. The consensus model showed high accuracy and high specificity of over 0.9. Comparison of the reactivity class predicted by the consensus model and the skin sensitization potential for humans revealed that all chemicals classified into the “Moderate-High” class were human skin sensitizers. In conclusion, the consensus model we constructed here may be a promising in silico screening tool to predict cysteine reactivity measured by the DPRA and skin sensitization potential of chemicals.