Data is everywhere. The present human lifestyle relies heavily on data. Machine learning is a huge domain that strives hard continuously to make great things out of the largely available data. With data in hand, a machine learning algorithm tries to find the pattern or the distribution of that data. Machine learning algorithms are usually defined and derived in a pattern-specific or a distribution-specific manner. For instance, Logistic Regression is a traditional machine learning algorithm meant specifically for a binary classification problem. Linear Regression is a traditional machine learning algorithm meant for the data that is linearly distributed in a multi-dimensional space. One specific algorithm cannot be applied for a problem of different nature.

To this end, Maximum Likelihood Estimation, simply known as MLE, is a traditional probabilistic approach that can be applied to data belonging to any distribution, i.e., Normal, Poisson, Bernoulli, etc. With prior assumption or knowledge about the data distribution, Maximum Likelihood Estimation helps find the most likely-to-occur distribution parameters. For instance, let us say we have data that is assumed to be normally distributed, but we do not know its mean and standard deviation parameters. Maximum Likelihood Estimation iteratively searches the most likely mean and standard deviation that could have generated the distribution. Moreover, Maximum Likelihood Estimation can be applied to both regression and classification problems.

Therefore, Maximum Likelihood Estimation is simply an optimization algorithm that searches for the most suitable parameters. Since we know the data distribution a priori, the algorithm attempts iteratively to find its pattern. The approach is much generalized, so that it is important to devise a user-defined Python function that solves the particular machine learning problem.

#developers corner #likelihood #log likelihood #maximum likelihood estimation #mle #probability distribution #python #regression #statistics