As the importance of email increases, the amount of malicious email is also increasing, so the need for malicious email filtering is growing. Since it is more economical to combine commodity hardware consisting of a media server or PC with a virtual environment to use as a single server resource and filter malicious email using machine learning techniques, we used a Hadoop MapReduce framework and Naïve Bayes amid various machine learning methods for malicious email filtering. Naïve Bayes was selected because it is one of the top machine learning methods (Support Vector Machine (SVM), Naïve Bayes, K-Nearest Neighbor (KNN), and Decision Tree) in terms of execution time and accuracy. In addition, to shorten the execution time by configuring the Hadoop framework using PC-based commodity hardware, among more than 200 Hadoop parameters, 21 major Hadoop configuration parameters affecting the execution time were selected and optimized through several cases of experiments. The execution time was shortened by setting the optimized values to the selected 21 parameters. The special thing is that in a distributed environment, increasing the block size for storing and managing data does not necessarily shorten the execution time and increasing the number of replicated data set does not increase proportionally the execution time. The types of main Hadoop configuration parameters include setting block size, adjusting the number of server threads, adjusting the number of blocks to be replicated, setting the number of MapReduce tasks, setting the memory size and heap size required for setting up MapReduce, adjusting the number of virtual cores allocated to the container, compressing the results of all MapReduce jobs running on the cluster, etc. The set value of the Hadoop configuration parameter was optimized to shorten the execution time through the experimental technique, based on the Hadoop framework, malicious mail prediction was made using the MapReduce programming technique using Naïve Bayes technology, a supervised machine learning technique and a bare metal server environment that does not apply the Hadoop framework, a prediction method for judging spam was also implemented with a Python program method with Naïve Bayes technology. Laplace smoothing was applied for stable calculation when calculating the prediction probability, and Laplace smoothing was applied to the Hadoop MapReduce Naïve Bayes method and the Bare metal Python Naïve Bayes method to compare the prediction error rate and accuracy between the two methods. To apply Laplace smoothing, Hadoop MapReduce Naïve Bayes method applied numerical value (1), Bare metal Python Naïve Bayes method applied eleven types of numerical values, among them, the result of applying 0.0001, which had the best duration time and good prediction error rate and accuracy, was compared with the result of Hadoop MapReduce Naïve Bayes method. As a result of comparing the accuracy and predictive error rate of the two methods, the Hadoop MapReduce Naïve Bayes method improved the accuracy of identifying spam and ham by 1.04 times and the prediction error rate by 6.04 times compared to the bare metal Python Naïve Bayes method. In conclusion, the Hadoop MapReduce Naïve Bayes machine learning method presented in this paper was able to shorten the execution time through the appropriate selection of Hadoop configuration parameters and optimized settings, and improved the system stability through easy spanning-out and distributed data set replication. Therefore, I think that the proposed system architecture and algorithm of this paper improved the accuracy of malicious mail prediction and reduced the prediction error rate.