0

I have a gene expression dataset for which I have the original data values and I have put this dataset through a few methods to test their accuracy of prediction. I need to compare the results from these methods to the original values. The points of comparison are not time dependent, for now I have used mean squared error rate for accuracy testing and scatterplot for the visualization(as in figure below)here the accumalation of the points to the middle signifies the accuracy of the method.

I would like to know how else to visualize the comparison or what else can be done with the dataset to prove the accuracy of the data. I am using an R package called 'CellMix' for processing the data and the link to the dataset is belowGene Expression dataset

Venkat
  • 3
  • 3
  • Note that it doesn't look like your residuals are homoscedastic. That might or might not be a problem. – Roland May 30 '16 at 07:00

1 Answers1

1

These are some of the statistical measures you could use:

  • Mean Squared Error (MSE)

  • Mean Absolute Error (MAE)

  • Sum of Squared Error (SSE)

  • Sum of Absolute Error (SAE)

Each of these measures can give you one perspective. For instance, while MSE gives you the average errors (or average deviations of predictions from original data), MAE is better in exhibiting the extreme points. Here is one sample explanation.

You can also use Supervised learning techniques:

  • Self-Organizing Maps (SOM) to cluster your data and analyze the data in each cluster to get to know your data better. If you have access to MATLAB, with your labeled data, it is very easy to do this. You can obtain nice illustrations as well.

  • Neural networks. With this one, MATLAB also gives you a measure of accuracy of the model and predictions in the end.

Community
  • 1
PM0087
  • 351
  • 2
  • 11
  • Hi PeyM87, Thank you so much for your reply, but can you elaborate a bit about how I can use SOM for comparing the data points from the sample but one value would be the original value and the other from the method. As I understand SOM would cluster the data, which would help in knowing anomalies but not sure how it would help with data comparison. Thanks – Venkat May 29 '16 at 15:46
  • You use SOM on your original data. You learn about your data and after you perform some analysis, your regression for instance, you know how well your model is performing intuitively. In addition, you can use the predicted data with SOM and compare the results with when you used original data with SOM. If they are the same, your model most probably does a good prediction. – PM0087 May 29 '16 at 15:51