{ "fetched_at": "2026-10-19T17:36:08Z", "source_url": "https://www.kaggle.com/competitions/2024-travelers-ness-statathon/data", "pages": [ { "name": "data-description", "content": "## Data Description\n\nThe data provided to you and your team consists of variables describing customers that asked for quotes. There are three datasets: at the policy, driver, and vehicle levels. Each row of the policy dataset corresponds to a single policy, which may be associated with multiple drivers and vehicles. Likewise, each row of the driver dataset represents a single driver, and each row of the vehicle dataset represents a single vehicle. It's easy to check that each driver and each vehicle are associated with just one policy. You may assume that all members on a policy live at the same address.\n\nYour company was only able to convert a fraction of the policies found in this sample. The policy dataset also has a training and test split variable called split. Note that the conversion indicator (the response variable) is missing for policies in the test split. Your task is to build a model on the training data and apply your model to predict the conversion indicator for each policy in test data. \n\n### Policies.csv\npolicy_id:\t\t Unique customer identifier
\nQuote_dt:\t\t Date the quote was submitted
\nquoted_amt:\t\t Quote amount (US dollars)
\nPrior_carrier_grp: Prior carrier group
\nCov_package_type: \t Level of coverage needed
\ndiscount:\t\t Whether or not a discount was applied to the quote amount
\nnumber_drivers:\t\t Number of drivers
\ncredit_score:\t\t Credit score of primary policy holder
\nnum_loaned_veh:\t Number of vehicles on policy that have a loan associated with them
\nnum_owned_veh:\t Number of owned vehicles on the policy
\nnum_leased_veh:\t Number of leased vehicles on the policy
\ntotal_number_veh:\t Total number vehicles on the policy
\nprimary_parking:\t Where car(s) are primarily parked
\nCAT_zone:\t\t Catastrophe risk zone
\nHome_policy_ind:\t Does customer has existing home insurance policy with Peace of Mind
\nzip:\t\t\t US zip code of policy holder
\nstate_id:\t\t State of policy holder
\ncounty_name:\t\t County of policy holder
\nAgent_cd:\t\t Unique agent code (8 digits)
\nsplit:\t\t\t Train/Test split
\n**convert_ind: \t\t Conversion indicator (0=no, 1=yes). This is the response variable**
\n\n### Drivers.csv:\npolicy_id:\t\t Unique customer identifier
\ngender:\t\t\t Gender of driver
\nage:\t\t\t Age of driver
\nhigh_education_ind: \t Higher education indicator
\nsafty_rating: \t\t Safety rating index of driver
\nliving_status: \t\t Driver’s living status (levels = ‘own’, ‘rent’, ‘neither’)
\n\n### Vehicles.csv:\npolicy_id:\t\t Unique customer identifier
\ncar_no:\t\t\t Unique car identifier (per policy)
\nownership_type:\t\t Whether the car is loaned, owned or leased
\ncolor:\t\t\t Vehicle color
\nage:\t\t\t Vehicle age
\nmake_model:\t\t Make and model of the vehicle\n\n# NOTE\n**Training and Test sets are combined across the files, and need to be separated manually by your group!**", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Evaluation", "content": "# Team Names\nMake sure to enroll with the same team names that you submitted in NESS registration!\n\n# Scoring \nThe evaluation metric for the competition is the *Area Under the Receiver Characteristic Operating Curve* (AUROC). Let TPR denote the True Positive Rate, and x denote the False Positive Rate for a classifier. Note that TPR is a function of x.
\nThen AUROC (A) is defined as $$ A : = \\int_{0}^{1} TPR(x) dx . $$\nThus, AUROC is a cumulative measure of a classifiers ability to detect true positives as we vary the type I error rate (i.e., False Positive Rate). The closer the AUROC is to 1, the better is your classifier. Your goal is to maximize AUROC. \n\nThe AUROC metric can be computed in both python in R using the roc-auc-score function in sklearn.metrics. and the auc function in pROC, respectively.\n\n# Submission Format \n For policies indicated in the train/test split as test, submission files should have two columns - policy_id and conv_prob, which denotes the Policy ID, and the estimated probability of conversion.\nNote, that the **conv_prob is a probability (i.e. between 0 and 1), and not a predicted class label**.\n\nThe CSV file should look like:\n\n| policy_id | conv_prob |\n| --- | --- |\n| policy_5 | .2456 |\n| policy_34 | .8976 |\netc. \n\n## Presentations:\n\nThe top teams will be invited to give a presentation to explain their models and findings to a panel of judges and Travelers employees and winners will be selected at NESS. Some questions to consider in your presentation are:\n\n* What methods did you consider (you don’t have to actually try all the methods you considered; just ones that you think would work for this problem)?\n* What method did you choose in the end, and why?\n* How did you do your variable selection and feature engineering?\n* What other variables not in the data set do you think might be useful?\n* How could your model predictions be used in practice?\n* Any interesting methods / findings, that can go unnoticed?", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "rules", "content": "## Modeling\n\n•\tEach group can make at most 3 submissions per day\n•\tWork together within group, but not between groups\n•\tYou can build the model with any software \n•\tWork with your group members and ask for clarification from discussion section\n•\tYou are only permitted to use the data provided – third party data sources are not to be used\n\n## Presentation Instructions\n\n•\tWhat methods did you consider (you don’t have to try all these methods; just ones that you think would work for this problem)?\n•\tWhat method did you choose in the end, and why?\n•\tHow did you do your variable selection? \n•\tWhat other variables not in the data set do you think might be useful?\n•\tAny interesting methods / findings, that can go unnoticed?\n\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "abstract", "content": "#Theme 1: Auto Quote Customer Conversion Model\n \nTravelers Insurance Company is looking for expert modelers in their personal automotive insurance department. For personal auto, anyone that asks for an estimated price for a policy from your company (also known as a quote) will receive one. However, of those quoted, only a fraction will choose this company as their insurer (versus other companies they also received quotes from). Your team has been asked to build what’s known as a conversion model, with the goal of understanding the population of policies Travelers Insurance Company is most likely to write (a.k.a. issue or convert). In other words, what types of customers is your company writing over its competitors?\n \nYour goals in this competition are as follows:\n \n•\tIdentify quoted policies that your company will convert (a.k.a. issue)\n \n•\tUnderstand key characteristics of policies your company tends to write, as well as those they tend not to write (e.g. understand quoted policies with both high and low conversion rates)\n \n•\tProvide a recommendation on how this information could be leveraged at Travelers Insurance Company\n \n \n## Keywords\n \nPolicy – account associated with a given customer.\nQuote – Price given to a potential customer for an insurance policy.\nIssue/write/convert – If your company ‘writes/issues/converts’ a policy, it means the customer chose your company as their insurer.\nConversion rates – the percent of quoted policies your company issues\n\n## Timeline\n- Competition opens : April 4, 2024\n- Competition closes : May 8, 2024\n- Presentation Dates : May 21, 2024\n- Winners Announced: May 21, 2024\n\n## Contacts\n\nNESS: \n•\tShane Sacco: shane.sacco@uconn.edu\n•\tKun Chen: kun.chen@uconn.edu \n\nTravelers:\n•\tGazi Inkiyad : GINKIYAD@travelers.com\n•\tRunzhi Zhou : RZHOU2@travelers.com\n•\tLubing Wang: LWANG5@travelers.com \n\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false } ] }