A Journey on Data Analysis, Predictive Analytics, Data Mining.

Monday, February 3, 2014

Data Analysis Steps

After going through the overview of tools & technologies needed to become a Data scientist in my previous blog post, in this post, we shall understand how to tackle a data analysis problem.
Any data analysis project starts with identifying a business problem where historical data exists. A business problem can be anything which can include prediction problems, analyzing customer behavior, identifying new patterns from past events, building recommendation engines etc.

The steps for solving a data analysis problem can be shown as below:
Steps to Solve Data Analysis Problem

Identify Business Problem:
 “Define Problem statement”
This is the first step of analysis. Business identifies a problem and a problem statement with desired outcome is defined. In this stage, a Data Scientist should understand the problem statement, the domain knowledge of the problem. After thorough understanding of the problem statement, a Hypothesis will be proposed.

 Data Acquisition:
“Identify data sources”
As a second step, all the data sources related to the problem statement will be identified and pulled into a central repository. The data sources can vary from SQL databases to text files to csv files to online data. If the data size is large we may use Hadoop to pull, store & pre-process the data.

Process/Clean Data:
 “The accuracy of the results of analysis depends on the quality of data” 
Data Clean step is considered to be one of the very important phases in Data analysis. The accuracy of the analysis depends on the quality of data.
Few approaches:
  • Formatting the data as per the data analytical tools we use.
  • Missing data handling
  • Data Transformations like normalizing the data Identifying outliers & handling etc.
Exploratory Analysis:
 “Embrace the data visually before diving further”
The objective of this step is to understand the main characteristics of the data. This analysis is generally done using visualizing tools. Performing an Exploratory analysis helps us:
  • to understand causes of an observed event
  • to understand the nature of the data we are dealing with
  • assess assumptions on which our analysis will be based
  • to identify the key features in the data needed for the analysis
Graphical Techniques:Scatter plots, box plots, histograms
Quantitative techniques: Mean, median, Mode, Standard deviation

Exploratory Analysis
Model Generation & Validation:
“Select-Train-Evaluate” 
This step involves extracting features from the data and feeding them into the machine learning algorithms to build a model. Model is the solution proposed for the problem statement. This step involves: Model selection, model training and model evaluation.
Model selection: Based on the type of business problem we are dealing, a model will be built. For example,if the objective of the analysis is to predict a future event, we need to build a Regression model for prediction.
Model Training: After selecting the Model for the analysis, the entire dataset is divided into 2 parts – Training data & Test Data. 3/4th of the entire data will be fed as input to the Model Algorithms.
Model Evaluation: Once the model is built. The next step is to test the model & validate it. The data used for testing the model is the remaining 1/3rd of the dataset in the previous step.
Model Evalation
Visualize Results:
 "Show the results visually" 
This is the final step of Data analysis where the results of the model & problem solved will be presented generally in visual plots/graphs.
Few visualizing tools: d3.js, ggplot2, tableau.

Visualize Results
Please go through the tools/technologies , skill set required to learn Data Analysis here