# Simple Time Series Prediction

I have a data set like this. Here the first column is date, the second column is Temperature, third one is humidity, fourth and fifth column are two other boolean data. I have data of 6 years like this.

2010-01-01,25.6,59,0,1
2010-01-02,25.6,60,0,1
2010-01-03,24.2,45,1,1
2010-01-04,26.3,20,0,1
2010-01-05,26.2,17,0,1
2010-01-06,24.3,65,0,0
2010-01-07,23.1,50,0,1
2010-01-08,26.3,25,1,0
2010-01-09,26.6,23,0,1
2010-01-10,24.3,60,0,1


And the label (The variable I want to predict) for this data set is: (boolean)

0,0,0,1,1,0,0,0,1,0


Now I want to implement Machine Learning to predict from this data set for a future time frame. I have almost no knowledge on machine learning. I want to use python to do it. Which library or methodology will be best and easiest for this? And can I have a simple sample code?

• Is the data periodic? – El Burro Apr 21 '17 at 12:32
• Not always. But there is a pattern. – nsssayom Apr 21 '17 at 12:33

## 2 Answers

Your problem looks to me more like a classification problem than a time series problem. My suggestion: Split the date into several sub-variables (year, month, day, week-of-day (maybe). Then just use this and the other values as input for a classification algorithm. Ideally you try several. I can recommend sklearn for this (http://scikit-learn.org/stable/auto_examples/classification/plot_classifier_comparison.html)

One advice: Depending on the classification algorithm you need to first normalize your data. I find the following blog useful for entry-level problems (with code examples-- if you want to approach this from a time-series perspective he also talks about this) and documentation http://machinelearningmastery.com/blog/.

Hope that helps.

• Okay! I am going to try this now... – nsssayom Apr 21 '17 at 12:44
• It can be a bit overwhelming at the beginning - but this should be the right place if you run into more specific problems :). – El Burro Apr 21 '17 at 12:51

Welcome to the wonderful and sometimes intimidating world of machine learning! Building on El Burro's reply, I would start with a recurrent neural network, and put a logistic layer at the top.

Recurrent neural networks are times series models in machine learning. They take input (the value of a variable at each point in time), perform some transformations, and give an output (a class label, in your case, but they can also be used for sequence-to-sequence prediction).

One specifically relevant blog post for you might be sequence classification.