Published on 2021-12-08, by Saba Siddiquie
Business problems are the key components enabling the businesses to uncover potential issues and resolve them using analytics. The process to build an analytical model begins with defining the objective and scope enables to set a clear vision allowing the process to stay on track. The steps of building an Analytical model can be summarized below:
It involves identifying ‘What and Why’ of the project to understand the requirements and set objectives or goals. A team of Business analyst, technical analysts, and SMEs along with stakeholders participate in discussion of business problem to have a better understanding of the issue/requirement.
It includes the validating the requirements shared by business. This will involve discussions with SMEs and brainstorming the team to align the business objective with analytical approach. This involves determining gaps and eliminating inconsistencies between business and analytical needs of the model.
In this phase, data is gathered from multiple sources as described by business. The data is checked for any inconsistencies. The data is cleaned and transformed by performing activities like normalizing the data, accounting for missing values etc.
This step can be combined during data cleaning where a preliminary data analysis is conducted to identify the trends and seasonality in the data. This further helps us to determine the suitable model for our problem. For example: To predict the climate change over the years, a time-series model is an effective approach on the other hand when predicting the number of loan defaulter, a logistic or Classification tree model could be beneficial.
Once your data is sorted and initial trend analysis is performed, model can be chosen and created to train the data. Based on the type of problem, a model could be logistic, decision tree or KNN. Selection of a model depends upon the business needs and objective. We must also look for any limitations regarding model implementation with respect to business scenario.
The model created above needs to be tested on the test data set that was partitioned during the modelling. Metrics such as Recall, Precision, F-1 Score, Accuracy can be used to evaluate model’s performance. The validation is performed on a new data set to ensure the model is as good as it was on training data. A completely new data set can be used during validation, or a part of data can be kept untouched/unprocessed during data gathering to be used in this step.
An overview or architecture must be developed prior to implementation like user manuals, handbooks, training etc. when planning to implement the model. Once the model is implemented for use, it should be checked regularly for inconsistencies. With time technology changes and to make the model useful in a long run, regular updates should be made.
These are few pointers which have helped me to achieve my goal when working on my analytics project. Following these steps allowed me to stay on track and monitor the progress of my project without missing the deadlines and avoiding any deviations. The steps may vary depending on each business but gives a generic overview to start building a model to beginners. Overall, the journey to build an analytical model is overwhelming and requires proper planning to avoid getting lost amidst the process.
To have a look on my analytics project, please follow the given link: Credit card fraud-code.ipynb - Colaboratory (google.com)
The data for this model can be found at here