Machine learning expert
Budget: $10 – $30 USD
This data is about solar wind. There should be 80 measurements in each column. These measurements are time series. If you know RNA you can use time series. We have 11 attributes. We can concatenate them. If we concatenate them, it will be 1000 dimensions. If we concatenate them, we will run into sum issues. We have range of attributes. We must preprocess the data. We can use normalization. First plot the histogram. We must do many things for preprocessing: Dimensionality reduction.
When we do preprocessing, we can do min max pre processing or Gaussian standard normalization. First plot the histogram. If the histogram looks like uniform distribution, then we must choose min max. If histogram looks like Gaussian, then use standard normalization. We may even have multi-mode. You can use any packages. We need to separate some data to validate our model. We need to validate the model and come up with the performance metrics.
The output is delay. Delay also has some range. We must first understand the delay if it satisfies the normal distribution or not. We just have one value for the delay. The first set of data point has one value for delay and the second set of data point has one value for delay. If you do not preprocess the data, it will be like a 1000 dimension and gives you one number.
Do you want to treat this as a regression problem or classification problem? Anything is fine regression or classification. We can also use a hybrid model (regression and classification). In the end we want to minimize the error. If you use regression problem you want to minimize the min square. If you use classification then you want to minimize the negative log likelihood. In order to any real world data projects you have to do several things. Try to understand the data. Plot the histogram, try PCA to find the dimensions and plot the graph for PCA. Once you understand the data you can decide what kind of preprocessing you want to do.
You always need to divide the data set into three parts: The training, the validation and the testing. (Reshuffle the data to do this). You have to decide what kind of model you want to use (Decision tree, regression tree, Random forest, Naïve Bayesian or Neural networks to boost your performance).
When we do preprocessing, we can do min max pre processing or Gaussian standard normalization. First plot the histogram. If the histogram looks like uniform distribution, then we must choose min max. If histogram looks like Gaussian, then use standard normalization. We may even have multi-mode. You can use any packages. We need to separate some data to validate our model. We need to validate the model and come up with the performance metrics.
The output is delay. Delay also has some range. We must first understand the delay if it satisfies the normal distribution or not. We just have one value for the delay. The first set of data point has one value for delay and the second set of data point has one value for delay. If you do not preprocess the data, it will be like a 1000 dimension and gives you one number.
Do you want to treat this as a regression problem or classification problem? Anything is fine regression or classification. We can also use a hybrid model (regression and classification). In the end we want to minimize the error. If you use regression problem you want to minimize the min square. If you use classification then you want to minimize the negative log likelihood. In order to any real world data projects you have to do several things. Try to understand the data. Plot the histogram, try PCA to find the dimensions and plot the graph for PCA. Once you understand the data you can decide what kind of preprocessing you want to do.
You always need to divide the data set into three parts: The training, the validation and the testing. (Reshuffle the data to do this). You have to decide what kind of model you want to use (Decision tree, regression tree, Random forest, Naïve Bayesian or Neural networks to boost your performance).