Graz University of Technology

Question

Asked 10th Mar, 2018

# Which tool of Matlab (backpropagation) is best for tournament prediction?

I'm doing a mini research on predicting a certain game tournament outcome (Dota 2) using backpropagation. My data structure consists of tables of (11) team's match histories against the other 10 teams. Basically 1 set of tables for each team. The tables contains inputs of:

1. the opponent's team ID (integer value ranging 1 to 11)

2. the (5) heroes ID used by team A and (5) heroes used by team B (integer value ranging 1 to 114)

In total, the input has 11 elements.

The target tables contains values "1" if team A wins and "0" if team B wins.

I was wondering which Matlab tool is best for this problem. I had tried using the nnstart tool but I don't know how to set the tool to use backpropagation. Also I would like to know, should I normalize the data for this problem? Is binary sigmoid the correct activation function to use?

Thanks in advance.

## All Answers (3)

For MatLab maybe you could consider trying MatConvNet (http://www.vlfeat.org/matconvnet/) and use it on CPU mode. For the amount of data you are proposing there probably is no need to use a GPU.

Regarding prediction of Dota 2 match outcome, I would strongly suggest using more data for each match. In order for the network to learn stuff like which heroes counter eachother, you would need more features than which position they are played on each match. In any case, I would recommend normalizing the features before feeding them to the network and you can use a sigmoid function or a cross-entropy loss for classification (output will be class 0:lose or 1:win).

Finally, in GitHub you will find several projects that try to predict Dota 2 match outcome using deep learning and that can serve as inspiration for further research.

University of Technology Sydney - The Royal Society

I suppose, there are many different kinds of approaches for prediction of tournament outcome. Although I have not had experienced on this problem, I have done a bunch of prediction works on electricity price or dynamic battery behavior by ANFIS and ANNs that are really work. You could read some of them.

1 Recommendation

## Similar questions and discussions

How should we deal with the lack of training data in a Artificial Neural Network task?

- Kanhaiya Sharma

I want to work on a ANN for Antenna Design task, but the size of data set is small and there are only 41*10 samples for training , testing & Validation purposes. How to increase sample size by using Oversampling concept in matlab?

In following paper already increased sample size from 41 to 322.
i unable to understand how it works (Oversampling) if you have idea kindly explain me.

Thanks

Kanhaiya Sharma

## Related Publications

En esta investigación se presenta un método para el reconocimiento de personas en un canal telefónico. El cual se basa en el comportamiento de las Redes Neuronales Artificiales (RNA), en particular, sobre la arquitectura del Perceptrón Multicapa mediante el algoritmo de aprendizaje Backpropagation aunado con la aplicación de la teoría de las Wavele...

The more advanced science and technology from various disciplines, currently rainfall can be predicted by carrying out various empirical approaches, one of which is by using Artificial Neural Networks (ANN). This study aims to apply ANN with backpropogation algorithm in predicting rainfall. The research data used is BPS data of the transfer city. T...

Artificial Neural Network (ANN) technique was used in forecasting the
Jordanian stock prices. The algorithm was developed using a feed forward multi
layer neural network; the network was trained using back propagation
algorithm. Software was developed by using MATLAB to simulate the
performance and efficiency of the algorithm. Simulation was conduc...