Dear Nagu,

Briefly, it depends at the nature of your data. For example, if the main purpose of using machine algorithms to you is to predict specific variable from normal textual data then most appropriate one is to use Support Vector Model algorithm.

Normally, the accuracy for the model depends on the leverage value of precision and recall of the model, others uses RMSE/MSE, others uses F-score measurements. So it depends of what you want to predict and which case you want to evaluate (false positive, true positive, false negative).

Also, the model accuracy depends hugely at the features that you already extracted from the data where you use them to train your model. You will notice that some features have higher effect at your model when you measure the accuracy, so selecting which feature to extract from the data is critical to the level where a lot of researchers now are studying how to engineer features with deep understanding of the term of factor analysis.

Hope this helps you...

Dear Nagu,

Briefly, it depends at the nature of your data. For example, if the main purpose of using machine algorithms to you is to predict specific variable from normal textual data then most appropriate one is to use Support Vector Model algorithm.

Normally, the accuracy for the model depends on the leverage value of precision and recall of the model, others uses RMSE/MSE, others uses F-score measurements. So it depends of what you want to predict and which case you want to evaluate (false positive, true positive, false negative).

Also, the model accuracy depends hugely at the features that you already extracted from the data where you use them to train your model. You will notice that some features have higher effect at your model when you measure the accuracy, so selecting which feature to extract from the data is critical to the level where a lot of researchers now are studying how to engineer features with deep understanding of the term of factor analysis.

Hope this helps you...

The commonly accepted practice is to try a variety of ML algorithms with a variety of parameters each and discover which one gives the best results. Also, try a variety of pre-processing and feature selection approaches in this process. Having said that, some ML algorithms have been already observed to perform better on certain types of datasets. For example, as Raad Bin Tareaf stated above, SVM have been observed to perform well on text data. Deep neural nets are typically considered very precise.

Your question is very broad, so I will attempt an answer, which will also be broad. With regards to utilizing more accurate machine learning models for data analytics, the answer varies according to the problem you are trying to solve, and the dataset you are using.

If you are looking at correlation of independent variable(s) with an outcome variable, then you may start with linear regression on the dataset, and calculate the correlation coefficient to determine 'goodness of fit.' In many cases, linear regression is good enough 'accuracy' to get you the prediction you want according to the independent variable(s) you input. If you want more 'accuracy,' meaning a higher correlation coefficient, then you may consider moving on to more advanced regression models, such as polynomial regression. However, in your quest to become more 'accurate' and fitting more advanced models, you run the risk of overfitting, and this will be counterproductive to your future predictions. So becoming more 'accurate' has the risk of overfitting. Therefore, 'accuracy' should not be the only focus for modelling.

But this is just one example of where Machine Learning can solve problems, and we know that Machine Learning has many models, with each specific model having great depth and complexity where researchers spend their entire careers building and optimizing such models.

An overall review of Machine Learning:

3 basic areas of Machine Learning:

1) Regression

2) Classification

3) Clustering

Advanced ML Models:

*Association Rule Learning

*Reinforcement Learning

*Natural Language Processing

*Deep Learning

More advanced:

*Dimensionality Reduction- allow to visualize data if too many dimensions

Tools:

*R (visualization and simple problems)

*Python (Deep Learning)

*Julia (good ML libraries)

*Hadoop

*Docker

*Work Flows

Hope this helps.