Beginning Anomaly Detection Using

Figure 6-34.  Plotting the testing and predicted datasets

Figure 6-33.  Code to compute the threshold

Chapter 6   Long Short-term memory modeLS 

---

241

Figure 

6-35

 shows the code to classify a datapoint as anomaly or normal.

Figure 

6-36

 shows the code to plot the data points with respect to the threshold.

Figure 6-35.  Code to classify a datapoint as anomaly or normal

Figure 6-36.  Code to plot the data points with respect to the threshold

Chapter 6   Long Short-term memory modeLS 

---

242

Figure 

6-37

 shows the code to append the anomaly flag to the dataframe.

Figure 6-37.  Code to append the anomaly flag to the dataframe

Figure 

6-38

 shows the code to generate a graph showing the anomalies.

Figure 6-38.  A graph showing anomalies

Chapter 6   Long Short-term memory modeLS 

---

243

In above graph you can spot an anomaly around Thanksgiving Day, one around New 

Year Eve, and another one possibly on a snow storm day in January.

If you play around with some of the parameters you used, such as number of time_

steps, threshold cutoffs, epochs of the neural network, batch size, and hidden layer, you 

will see different results.

A good way to improve the detection is to curate good normal data, use identified 

anomalies, and put it in the mix to have a way to tune the parameters until you get good 

matches on the identified anomalies.