Problem Description : Predictive maintenance of machinery is very critical from both cost and safety perspectives. Cost ofserving/replacing a perfectly working part can be significant and can add huge financial burden to manufacturers and customers besides causing machine down time. On the other hand, cost of notmaintaining a failed part can result in human safety disasters. Thus, predicting events related to maintenance based on past events till failure helps avoid breakdown and achieve full useful life and thus availability of systems. The data on failures or trends about degradation of a machine over time can help in predicting a potential failure based on patterns and helps estimate time to a failure based on the progression of identified faults. Working backwards the data can help develop early warning signals to plan for maintenance which includes corrective actions. A locomotive manufacturing company observes failures in the engines of several locomotives of their fleet. The data captured through sensors and inspection of previous fleet is given. The data depicts the scheduled and unexpected services done for each machine component. Unexpected services include error, repair, replacement etc and related data is provided with time stamps. You are expected to create an analytical and modelling framework to predict the major possible future events/ action points for each MachineID ie., like any “ComponentRepair”(major), “ComponentReplacement”, “NoIssue” (Minor error or normal status) , in the next one month for each machine. And you are also required to obtain the actionable top 10 data insights (patterns) for each of “ComponentRepair”, “ComponentReplacement” class levels using the tree based algorithms. About Data: The data provides the information on the condition of the various components of the machine including the data collected through sensors. The data depicts the scheduled and unexpected services that took place for each machine as historical data. Every single machine is observed at various time stamps, to record the details of its operating condition like any error, repair, scheduled maintenance, replacement etc for each component of that specific machine. Objective: You are expected to create an analytical and modelling framework to predict the future events/action points based on known past events/action points for each MachineID ie., what would be the possible major action point, like any “ComponentRepair”, “ComponentReplacement”, “NoIssue” , in next one month for each machine. And also obtain the actionable top 10 data insights (patterns) for each of “ComponentRepair”, “ComponentReplacement” classes using the tree based algorithms.

H2O_GBM_Tuning

Random Forest has been used to predict whether a bank would go bankrupt or not

Predicting Labour Wages using Ridge and Lasso Regression. Ridge and Lasso Regression RSS(β)+λ∑j=1pβ2jRSS(β)+λ∑j=1pβj2 RSS(β)+λ∑j=1p|βj|