Conceptual Machine Learning Framework for Initial Data Analysis

This century has witnessed the emergence of new branch of science—data science that facilitates the analysis of large amounts of data which in turn helps in taking model-based data-driven decisions. The prelude to any successful analytical model building and implementation phase is a properly conducted initial data analysis stage. IDA encompasses laborious tasks of data cleansing: missing value treatment, outlier detection, checking the veracity of data, data transformation, and thus preparing data for model building. A systematic, disciplined, and non-personalized approach to IDA reduces the probability of incorrect and inaccurate results from the model. The amount of data presented for model building today makes the IDA stage a very crucial task which cannot be manually conducted. Machine learning can be applied to analyze complex and bigger data, find patterns accurately, etc. Hence, it could also be used for data preparation prior to model building. This paper tries to reduce the ad hoc nature of IDA by providing a conceptual framework using machine learning.