Human-vs-Bot

This is the first part of the Human-vs-Bot project. Here, we simulate data to mimic plausible behavioral differences between human and bot users—focusing specifically on response timing patterns. We walk through the functions used to generate and clean the data, with the goal of producing a dataset suitable for classification modeling.

Data Simulation and Preparation

The functions we use generate synthetic user session data, but the raw output isn’t immediately ready for modeling. In this section, we explain how the simulated data is structured and describe the key steps we took to clean and prepare it for analysis.

There are four core functions in this process:

The _startTimes functions create session identifiers and assign start times for each user, while the _timesDistributionDF functions simulate time-stamped responses. Human response intervals are drawn from a normal distribution, while bot responses follow an exponential distribution.

Start Time Generation

The _startTimes functions are structurally similar for both humans and bots. Each function assigns a unique user ID and generates session IDs based on simulated start times.

Response Time Distributions

Humans are simulated with more variable and normally distributed response intervals, while bots produce faster, more consistent responses following an exponential distribution. These assumptions are deliberately simple—they are not meant to reflect real-world bot detection methods but to illustrate how even basic behavioral patterns can be used to construct predictive models.

Feature Development

The functions above generate data with four key columns:

Each id and session combination has multiple response time entries. The type (bot or human) is also constant per session. From this, we derived five additional features. Four are used in classification models, and one supports visualization. One key feature is the number of responses per session. While an earlier version referred to this as clicks_per_session, we updated the terminology to responses to better reflect varied input behavior across users.

  • responses_per_session
  • duration
  • rate
  • rate_10min
  • inter_response

    • While earlier drafts referred to clicks_per_session, we transitioned to responses_per_session to better capture varied user input—such as mouse movement, screen touches, or scrolling. This broader definition reflects the kind of behaviors we'd expect in realistic sessions and leaves room for richer analysis in future iterations.

    responses_per_session was determined by the number of unique times, or rows per id and session_id. duration was determined by ordering the rows with respect to the time column and calculating the distince between the last time and first time by id and session_id. The remaining feature calculations can be found in the R script simulateData_andPrepUpdated.R.

    Results

    Histograms

    Scatter Plot