In this paper, we use the high frequency trading data of New York Stock Exchange to analyze the bid-ask spread components. It is found that there is an exponential relationship between the log returns of quoted midpoints and the trade volume. We also observe a negative linear correlation between the changes of quoted depth and the trade volume. Furthermore, changes of the quoted ask depth and the quoted bid depth are asymmetric due to the trading direction. Furthermore, statistical quality control charts, p-charts, are built for fixed number of trades to monitor unusual trades entering the market. Finally, logistic regression models are established to predict the probabilities of unusual trades entering the market based on the quotes and the quoted depth adjustments of the market makers.