# Stochastically find dissimilarity between time series data and model prediction

• willfurnass
In summary, stochastically finding dissimilarity between time series data and model prediction is a method used to evaluate the accuracy of a model in predicting future values based on historical data. It involves comparing the model's predicted values with the actual values in a randomized manner, using metrics such as mean squared error, mean absolute error, correlation coefficient, and root mean squared error. This evaluation can help improve the model by identifying discrepancies and weaknesses, but it is limited by the quality and quantity of data available and may not be suitable for all types of time series data.
willfurnass
I have many time series, with each featuring fifteen data points (x,t). I wish to fit a parameterised model to each data series. At present I'm using particle swarm optimisation (PSO) for this purpose. Within my objective function I quantify fitness using the Bray Curtis distance between a) the data series and b) the prediction corresponding to a candidate solution.

The problem I have is that the PSO fitting does not presently account for uncertainty in either x or t. The variances in x and t are most likely themselves invariant and I'd ideally like to treat each data point as a 2D Gaussian PDF.

My question is this: could Gaussian Processes (GPs) be used within my objective function to _stochastically_ determine the dissimilarity between a 15-point data series and the 15 (or more) point prediction corresponding to a candidate solution?

Regards,

Will Furnass

Dear Will,

Thank you for sharing your approach to fitting parameterised models to time series data. It sounds like you have a well-defined objective function and are using an effective optimization method in PSO.

In regards to accounting for uncertainty in your data, Gaussian Processes (GPs) could certainly be used within your objective function to determine the dissimilarity between your data series and predictions. GPs are a powerful tool for modeling and predicting data, particularly in cases where there may be uncertainty or noise present. They can also be used for regression and optimization tasks, making them a good candidate for your problem.

One potential approach could be to use GPs to model your data series and then use the predicted values from the GP as a comparison to your candidate solutions. This would allow you to incorporate the uncertainty in your data into your objective function and potentially improve the performance of your optimization.

However, it's important to note that using GPs in this way may also introduce additional complexity and computational overhead. You may need to carefully consider the balance between incorporating uncertainty and maintaining the efficiency of your optimization method.

Overall, I think using GPs in your objective function is a promising idea and could potentially lead to improved results. I would recommend exploring this approach further and conducting some experiments to see how well it performs compared to your current method.

Best of luck with your research!

Regards,

## 1. What is the purpose of stochastically finding dissimilarity between time series data and model prediction?

The purpose of stochastically finding dissimilarity between time series data and model prediction is to evaluate the accuracy of the model in predicting future values based on historical data. It helps identify any discrepancies between the model's predicted values and the actual values, which can be used to improve the model and make more accurate predictions in the future.

## 2. How is stochastically finding dissimilarity between time series data and model prediction different from traditional methods?

Stochastically finding dissimilarity involves comparing the model's predicted values with the actual values in a randomized manner, rather than simply calculating the difference between the two. This allows for a more comprehensive evaluation of the model's performance, as it takes into account the variability and randomness in the data.

## 3. What are the key metrics used in stochastically finding dissimilarity between time series data and model prediction?

The key metrics used in stochastically finding dissimilarity include mean squared error, mean absolute error, correlation coefficient, and root mean squared error. These metrics provide a measure of the difference between the model's predicted values and the actual values, and can help identify areas for improvement in the model.

## 4. How does stochastically finding dissimilarity help in improving time series models?

By identifying discrepancies between the model's predicted values and the actual values, stochastically finding dissimilarity can help in identifying the weaknesses in the model and improving its performance. This can involve adjusting the model parameters, incorporating new data, or using a different modeling approach.

## 5. What are the limitations of stochastically finding dissimilarity between time series data and model prediction?

Stochastically finding dissimilarity is dependent on the quality and quantity of data available. If the data is limited or of poor quality, the evaluation of the model's performance may not be accurate. Additionally, this method may not be suitable for all types of time series data, and may not provide insights into the underlying patterns and trends in the data.

Replies
14
Views
969
Replies
1
Views
1K
Replies
5
Views
2K
Replies
2
Views
1K
Replies
1
Views
2K
Replies
24
Views
2K
Replies
13
Views
4K
Replies
12
Views
2K
Replies
1
Views
2K
Replies
8
Views
1K