Explainable and Interpretable Forecasts on Non-Smooth Multivariate Time Series for Responsible Gameplay

TL;DR

This work tackles the need for explainable forecasting in non-smooth, multi-source time series arising from online gameplay, where predicting overindulgence must be accompanied by human-understandable rationale. It introduces the Actionable Forecasting Network (AFN), an integrated framework that combines a Transition Module with a Deep Markov Model, a Conditional VAE with Latent SOM (ConVAE-SOM), and an Intelligent Forecasting Module (LSTM with Attention and a Damping Factor Network) to produce accurate forecasts, smooth interpretable trajectories, and explanations via SHAP over SOM clusters. AFN achieves a notable $25\%$ MSE improvement over SOM-VAE on player data and provides mechanisms to attribute unfavorable trajectories to specific future time steps, enabling personalized interventions; it also demonstrates the ability to detect at-risk players up to $\approx4$ weeks in advance for proactive support. The proposed system is validated on real-world gaming data and shows practical impact by enabling early risk mitigation, interpretable visualization of risk trajectories, and a path toward responsible gameplay through targeted, time-aware interventions.

Abstract

Multi-variate Time Series (MTS) forecasting has made large strides (with very negligible errors) through recent advancements in neural networks, e.g., Transformers. However, in critical situations like predicting gaming overindulgence that affects one's mental well-being; an accurate forecast without a contributing evidence (explanation) is irrelevant. Hence, it becomes important that the forecasts are Interpretable - intermediate representation of the forecasted trajectory is comprehensible; as well as Explainable - attentive input features and events are accessible for a personalized and timely intervention of players at risk. While the contributing state of the art research on interpretability primarily focuses on temporally-smooth single-process driven time series data, our online multi-player gameplay data demonstrates intractable temporal randomness due to intrinsic orthogonality between player's game outcome and their intent to engage further. We introduce a novel deep Actionable Forecasting Network (AFN), which addresses the inter-dependent challenges associated with three exclusive objectives - 1) forecasting accuracy; 2) smooth comprehensible trajectory and 3) explanations via multi-dimensional input features while tackling the challenges introduced by our non-smooth temporal data, together in one single solution. AFN establishes a \it{new benchmark} via: (i) achieving 25% improvement on the MSE of the forecasts on player data in comparison to the SOM-VAE based SOTA networks; (ii) attributing unfavourable progression of a player's time series to a specific future time step(s), with the premise of eliminating near-future overindulgent player volume by over 18% with player specific actionable inputs feature(s) and (iii) proactively detecting over 23% (100% jump from SOTA) of the to-be overindulgent, players on an average, 4 weeks in advance.

Figures (10)

• Figure 1: AFN Architecture
• Figure 2: Trajectory of Player 1 and Player 2 on a 2D SOM grid of risk-related clusters. AFN forecasts the trajectory of Player 1 towards the darker(risky) clusters, while it forecast a smooth trajectory on the lighter(healthy) clusters for Player 2. These maps enables the interpretability by visualising the state of player at each time step.
• Figure 3: Explanations of final forecast for Player 1 and Player 2 by showing the respective Attention Point and Dominant Feature respectively.
• Figure 4: heatmap of Dominant and a Non-Dominant Feature. Dominant feature clearly shows higher Shapley magnitude throughout the attentive period for the player demonstrated in Figure \ref{['fig:exp_trajec']}
• Figure 5: Illustration of how a forecasted SR trajectory for a player can be restored to a SH via timely and personalized intervention