In this technical paper, we propose a machine learning-based approach for predicting customer claims in hybrid fiber coaxial (HFC) subscribers. We are currently finalizing a proof of concept of the model in some areas of our network. The proposed approach involves collecting data from multiple sources such as network logs, customer service records, hourly collected information from over 3.5 million cable modem (CM) data over cable service interface specification (DOCSIS) and others. We have been able to overcome many technical challenges, two of the most difficult tasks have been dealing with an extremely unbalanced dataset and label noise. We use several machine learning algorithms, and finally the one selected was extreme gradient boosting (XGBoost) to build the ML model. As a result, we obtain a daily list of customers with high probability to start a claim (95%). From this list, with Customer Experience (CX) and Field Service teams, we can make proactive calls to solve customer problems remotely or to send a technician to their home if necessary. The proposed approach can help HFC network service providers to proactively identify potential issues in their subscribers' connections and take preventive measures to avoid customer claims that end up in the generation of a technical service ticket. This can lead to improved customer satisfaction, reduced churn rates, and lower operational costs. We measured through surveys how was the experience of our customers and we found that this proactive action has a positive impact on their satisfaction. Furthermore, the approach can be extended to other types of networks and where predictive maintenance and customer experience management are critical.