Revenue Forecast & Event Classifier
Budget: ₹100 – ₹400 INR
I’ve got a collection of time-stamped web server logs and I want to squeeze two clear outcomes from them:
1. A reliable time-series model that forecasts our revenue day-to-day (and ideally beyond) so we can plan inventory and campaigns with confidence.
2. A companion classifier that flags and categorises IT events hidden in the same log stream—anything from routine spikes to anomalies that hint at trouble—so operations can react before customers notice.
The data is already centralised; you’ll receive the raw log files plus a cleaned-up sample to speed exploration. I’m open to the modelling stack you prefer—Python with Prophet, ARIMA, LSTM, or even Facebook’s NeuralProphet are fine—as long as the forecasts are explainable and the event labels can be exported to our monitoring dashboard.
Key points I need to see:
• End-to-end notebook or script that ingests the logs, performs feature engineering, trains the revenue forecasting model, and outputs future revenue with confidence intervals.
• A supervised (or semi-supervised, if you argue the case) classifier for IT events, complete with training routine and evaluation report (precision, recall, F1).
• Clear README covering setup, dependencies, and how to retrain with fresh logs.
• One-page summary of insights: top drivers of revenue changes and the most frequent event types discovered.
Push everything to a private Git repo; I’ll test it against a withheld slice of data. If forecasts track within an acceptable MAE and the classifier reaches the agreed F1 score, we’ll call it done and move on to rollout.
1. A reliable time-series model that forecasts our revenue day-to-day (and ideally beyond) so we can plan inventory and campaigns with confidence.
2. A companion classifier that flags and categorises IT events hidden in the same log stream—anything from routine spikes to anomalies that hint at trouble—so operations can react before customers notice.
The data is already centralised; you’ll receive the raw log files plus a cleaned-up sample to speed exploration. I’m open to the modelling stack you prefer—Python with Prophet, ARIMA, LSTM, or even Facebook’s NeuralProphet are fine—as long as the forecasts are explainable and the event labels can be exported to our monitoring dashboard.
Key points I need to see:
• End-to-end notebook or script that ingests the logs, performs feature engineering, trains the revenue forecasting model, and outputs future revenue with confidence intervals.
• A supervised (or semi-supervised, if you argue the case) classifier for IT events, complete with training routine and evaluation report (precision, recall, F1).
• Clear README covering setup, dependencies, and how to retrain with fresh logs.
• One-page summary of insights: top drivers of revenue changes and the most frequent event types discovered.
Push everything to a private Git repo; I’ll test it against a withheld slice of data. If forecasts track within an acceptable MAE and the classifier reaches the agreed F1 score, we’ll call it done and move on to rollout.