Page 144 - Building Big Data Applications
P. 144

142   Building Big Data Applications


                  Macroenvironment variablesdMacroenvironment variables focus on identifying
                   changes in the world that could affect the customer.
                  The churn variable that we will use for our application and dashboard include
                   - Demographic variables and behavioral variables
                   - All other variables that are available in the bank database
                Now the definition of churn and associated modes of activities are clear, we need to
             define how we will engage in tracking the engagement windows and its outcomes.
             Engagement windows for customer activity include
               Online banking
               Internet shopping
               Wire transfers
               Call center activity
               Campaign response
               Computer banking
               Inebranch banking
                Each of these activities generate several logs for customer, activity, channel, time of
             engagement, model of activity, sentiment expressed across social media, and influencer
             and follower models of activity. Additional data can be purchased today for similar ac-
             tivity across competitors and in regulatory compliance, the actual username and sen-
             sitive data if any are not shared. The next step is to identify the algorithms that can
             process this data independently and can harness and tether data once processed to align
             and provide insights as needed by the users.
                Algorithms to be used in artificial intelligence and machine learning models will
             include the following

               Naı ¨ve Bayes Classifier Algorithm
               K Means Clustering Algorithm
               Support Vector Machine Algorithm
               Apriori Algorithm
               Linear Regression
               Logistic Regression
               Artificial Neural Networks
               Random Forests
               Decision Trees
               Nearest Neighbors

                The models will be built using TensorFlow, Caffe 2.0, and Keras models, with several
             integration points of data being transformed and delivered with the models. The
             analytical outcomes will plot the results based on the input data and create graphs. The
             results typically will include outcomes that will provide identities of customers who are
             likely to churn. To provide more details on the reasons and the linage of data, the model
   139   140   141   142   143   144   145   146   147   148   149