Planning and Modelling

This week at ISM, I continued the planning phase of my final product on the basis of creating a model to predict insurance claims. During my meeting with Mr. Cambpell, we established a plan to initiate the creation of a model for the insurance algorithm. The first step to the process was finding and modelling the data necessary to train the algorithm. In the case of my original work, I had to receive a large amount of samples from other people and their handwriting in order to train the algorithm effectively. However, due to technical limitations, I had to use a large amount of samples from single people and couldn't access the varied data necessary to create an effective, accurate, and adaptive algorithm that can be applied to a wide variety of scenarios. In the case of insurance, this aspect of algorithm development is even more important in order to ensure fair and legitimate rates for the client. Ultimately, Mr. Campbell and I concurred that the best way to develop this algorithm was through reinforcement learning. In order to facilitate this learning, we need massive amounts of datasets in an organized and cleaned up manner. The best resources that we found for this were Kaggle and a UC Irvine public database. Over the course of the next week, I plan to implement and organize this data into a model using a data modelling software. Mr. Campbell and I discussed two different softwares to use for the future, TensorFlow and Jupyter Notebooks. While I used TensorFlow for my last project, the monolithic nature of the software makes it quite heavy and inaccessible for the average user. Ultimately, I learned my lesson from the last project and want to focus more on the creation of the AI algorithm, not the modelling of the data. Thus, I intend to follow Mr. Campbell's advice of using Jupyter Notebooks, a much simpler and more streamlined data analysis tool that they use at Toyota. Over the next week, I can use Jupyter to find and analyze the data necessary for the next step of my final product, the actual AI development.