The csv file is in data folder. It has data for 16 CPUs of machine 1. The CPU,RAM, disk utilization and other facors for 5 days
are mentioned. 
The three jupyter notebooks need to be executed in order viz final_df_1,mac_1,mac_1predict. 

final_df_1.ipynb file is for the following :
The parameters I have taken into account before applying PCA are: 
1) CPU utilization(CPU001 to CPU016)
We can just add the Usr%, Sys% and Wait% to get the total CPU utilization for each CPU.(ignoring the idle%) We can then take the log for each CPU and add all such values to get a total value for 16 CPUs. Taking the log will help us to normalize the score.
2) RAM utilization(MEM)
	We can keep 2 parameters in this:
	Swap space unused : swaptotal- swapfree+ Inactive
	RAM unused = memtotal-(memfree+buffers+cached)
 	Combine these two parameters by adding them and thus getting a consolidated value for each time stamp.
3) Hard disk utilization(DISKBSIZE,DISKBUSY(%),DISKREAD,DISKWRITE,DISKXFER)
i)We calculate the disk size in use by multiplying the % of DISKBUSY WITH DISKBSIZE for each drive and then add all to get for 
a particular timestamp.
ii)We also add the DISKREAD,WRITE for all hard disks and divide it with the sum of all DISKTRANSFER at a particular timestamp.
Divide DISKSIZE(i) By the computed result in step(ii) 
4) NetPackets
The ratio of lo-read-KB/s and lo-write-KB/s should be one i.e values of both the terms must be equal. For eth1-read-KB/s and
eth1-write-KB/s we will add the values and the outlier value will   
make a difference
PCA is then applied to reduce these 4 value to a single value for each timestamp.

	
mac_1.ipynb notebook is used to train a LSTM model. We use univariate time series forecasting as we have already reduced it to 
a single feature using PCA stated above.
The brief approach used:
1. A moving forward window of size 50, which means we will use the first 50 data points as out input X to predict y1 — 51st data 
point. Next, we will use the window between 1 to 51 data points as input X to predict y2 i.e., the 52nd data point and so on.
Here is the plot of first 50 data points 
2. use a two layered LSTM architecture coupled with a dense output layer to make a prediction.
3. We will look at couple of approaches to predict the output — a) Forecasting step by step on the test data set
b)Feed the previous prediction back into the input window by moving it one step forward and then predict at the current time step.
Preparing the 3D input vector for the LSTM.The input vector for LSTM is 3D array: (num_samples, num_time_steps, num_features)
Save the weights for the model trained (can be found in weight folder), plot the predicted values to visualize the trend in data.

mac_1predict.ipynb notebook predicts the reduced parameterized value for the next 20 timestamps after taking a particular
timestamp as input.

Just we need to divide the reduced values from the training set into 2 classes , Normal case and System Failure generation. 
We can use SVM to classify the value to either Normal or Failure class.