Abstract—Pancreatic cancer is one of the deadliest cancers due to the fact that it does not exhibit symptoms in the early stages. Furthermore, when pancreatic cancer gets diagnosed, it is usually too late. Consequently, early diagnosis is highly essential. The dawn of proteomics has brought with it a glimpse of hope of uncovering biomarkers that can be indicative of early pancreatic cancer. Proteome profiling techniques have become popular in the recent years to try to make sense of high-dimensional proteomic data and to find discrepancies between proteomes of healthy samples and cancerous samples. However, the high dimensionality of proteomics data coupled with small sample size poses a challenge. In this paper, we propose a framework using a hybrid logistic tree technique together with a feature selection technique to diagnose premalignant pancreatic cancer. We have validated our framework on a pancreatic cancer peptide mass spectrometry dataset. Satisfactory preliminary experimental results demonstrate the efficacy of our framework.