I am working on a indoor localization based on magnetometer.

I have 9 separate time-series datasets of sensor readings taken from coordinates 00, 01, 02, 10, 11, and so on until 22. Basically I am using my own coordinate system and gathered data. The coordinate system looks like this:

0,0 | 0,1 | 0,2

1,0 | 1,1 | 1,2

2,0 | 2,1 | 2,2

The dataset has columns timestamp, X, Y, Z and Magnitude. Here's what all my dataset looks like:


Edit: I collected a sensor readings data at one place with coordinate say 0,0 and saved the file as (coor00.csv) then at another place with coordinate say 0,1 and saved the file as (coor01.csv) and so on.

What I want to predict is coordinates. I want to build a simple classifier that predicts the coordinate/place/area based on sensor readings.

Since I don't have any labels/tags so I thought of creating one myself by simply adding a column named label. This column will be my target variable. So in coor00.csv file I will add a column label which will have details of what coordinates the sensr readings were taken from and I will do the same for other datasets. Then I will combine 9 datasets into a dataframe and then run a classifier after splitting the dataset into train and test.

There are plenty resources out there, but I just want to know how/where to start. I want to know if this is the right way.

I plan on using RandomForest classifier but I would appreciate any suggestions on what kind of classifier algorithms should be used?

  • $\begingroup$ Could you please elaborate a bit -- are the sensors observing the same events? i.e. are the 9 datasets describing the same process from different perspectives or there is some other relationship between the collected data points? What are you trying to predict exactly? What is the label column? $\endgroup$ – Vlad_Z Jul 29 '20 at 16:43
  • $\begingroup$ @Vlad_Z Please check the post again. I've made some edits. Let me know if you can help me out. $\endgroup$ – harry r Jul 29 '20 at 17:03

By looking at your datasets, it resembles a multivariate time series problem , not sure why you are opting for Random Forest classifier ?

  1. I would suggest you to start implementing simple statistical algorithm first here is the link

  2. Then explore on complex Deep Learning algorithm based on the data set size , please find the referenced link

Thanks, Durga

  • $\begingroup$ Okay Thank you @DurgaK I will refer the links. But what about the target variable. Is it right to just create our own labels and then use them as target variables or am I wrong? $\endgroup$ – harry r Jul 29 '20 at 18:26
  • $\begingroup$ Time series is part of regression , so i don't think target label were required for forecasting( Time Trends ), i would request you to first be clear with your expected outcome, i mean what you actually needs to achieve with the above dataset , then start exploring which type of algorithm suits to your requirements. Thanks. $\endgroup$ – Durga K Jul 29 '20 at 18:33
  • $\begingroup$ i would request you to first be clear with your expected outcome . i cannot stress enough that the outcome should be a coordinate i.e. 00 based on sensor readings. I want the classifier to predict the coordinate based on sensor readings $\endgroup$ – harry r Jul 29 '20 at 18:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.