I am working on a project that should predict Alarm based on the input data. I am trying to use a supervised learning algorithm. But I do not have the exact target value based on the input for feeding the learning algorithm. I am trying to generate the target value manually based on the input data (Current Target = previous input[frequency] +constant). But the plot of the target is not a smooth curve. A sample is given below.
In the graph given below the yellow line plots the current alarm (manually set), pink is the generated target alarm value and red is the input (frequency).
As the pink plot isn't a smooth curve, it cannot be used as a efficient target output for learning. I have also tried normalization of the graph which gives a really undesired output.
What method should be used to generate the target value (There are no fixed rule for the alarm value)?
How can the curve be smoothed desirably?
Working on Knime with a dataset of 6000 rows.