Earlier detection of prostate cancer (PCA) patients could improve patient outcomes through the increased scope to follow-up and treatment to the patient at high risk of PCA since healthcare professionals are struggling to correctly identify and successfully treating PCA cancer. To improve patient outcomes requires the prediction of patients at high risk correctly for proper clinical decision making. We, therefore, used a deep learning technique for accurate prediction of PCA patients one year earlier with minimal features from electronic health records. We retrieved data of 4,071 PCA patients from the Taiwan National Health Insurance database (NHID) between 1999 and 2013. Patients' age, sex, comorbidities, and medication history were used to developed and test a convolutional neural network (CNN) based prediction model. The area under the receiver operating curve for the prediction of PCA was 0.94. The sensitivity and specificity of CNN were 0.87 and 0.88, respectively. In this evaluation of data with minimal features from electronic health records, CNN had high sensitivity and specificity for identifying PCA. Although accurate and earlier detection of PCA is known to be challenging, our deep learning-based prediction approach may offer great benefits for correctly stratifying the patients at high risk of PCA that enables earlier decision making for proper treatments.