# References for longitudinal data analysis

So my goal is to study longitudinal data (data in time series) by applying some data mining techniques. Ultimately I want to be able to "predict" outcomes. For example, a study of patients along the years/months (repeated measures) can be used to predict some diseases using things like logistic regression, neural networks, etc..

I'm using R. However I'm having a hard time finding good references on this subject. I have a mathematic background, even though I've applied some data mining techniques before, I've never applied those to time series.

If someone can recommend me some good theoretical references or hands-on examples in R, that would be greatly appreciated. Also if you know a relevant article on time series analysis, that would be great too. Thanks in advance.

There are a lot of great resources on time series which I found really handy examples with hands-on examples in R. Though, I myself even as a regular python user find these below mentioned articles and blogs easy, to begin with, Time Series Analysis in R. My suggestion is to go through the statistical models in Time Series like ARMA, ARIMA, etc., and then proceed with neural networks.

Analytics Vidhya and Kaggle are great sources to look out in the beginning.

Check out the kernels on LANL Earthquake Prediction Challenge on Kaggle for more analysis on the longitudinal data

In data science, people normally refer to “longitudinal data” as one or more time series. So you can try searching for that. A classic is to use ARIMA (available in many languages including R). Personally I also like Facebook’s Prophet library as a starting point, because it’s easier to use and works surprisingly well. Also available in R.

• Thanks! In fact, I've searched and just realized that longitudinal/panel data is very similar to time series. Moreover, there is a lot more stuff about time series than longitudinal data. Sep 17, 2019 at 13:39

I have been working on a similar kind of problem and below is my methodology. Hope it is of help:

• Classify and analyze your time series problem. This will help narrowing down the domain of your problem and selection of technique. It can be done on the basis of factors including but not limited to:

1. Multivariate on univariate?
2. Regression or Classification?
3. Size of inputs (labels), outputs (features)?
4. Total number of examples?
• Read up on different statistical techniques for analyzing/forecasting time series data with a special focus on which problem could be solved by the factors you determined in previous step. Some techniques are ARIMA,Smoothing Average etc.
• If you feel that performance could be increased further, start experimenting with classical machine learning models. Some examples are Logistic Regression, SVMs, Decision Trees, Random Forests. You will have to read up on each for its pros and cons.
• If you still do not feel satisfied with the performance, you can try with deep learning and neural networks. Its better that you read up on a comprehensive resource like an online deep learning course etc.

A comprehensive resource I have found on time series and sequences is : Coursera course on Time Series by deeplearning.ai.