How can incorporate downward sloping granularity into my machine learning tensorflow.js algorithm?

I have a data set that I've plotted to show sales at time x over price y. I'm using a model to predict the price y over this time period.

However, it seems that no matter how much complexity I add or subtract, I can't get the model to include downward sloping gradients as part of its final output?

How might I achieve this?

An example output :

The model

function createModel() {
            const model = tf.sequential();
            
            //adding a layer
            model.add(tf.layers.dense({
                units: 5, //number of nodes in layer
                useBias: true, //adds a bias parameter 
                activation: 'tanh', //activation function
                inputDim: 1, //amount of inputs
            }));

            //not an input layer so has no inputDim property
            model.add(tf.layers.dense({
                units: 10, //number of nodes in layer
                useBias: true, //adds a bias parameter 
                activation: 'softmax', //activation function
            }));

            model.add(tf.layers.dense({
                units: 10, //number of nodes in layer
                useBias: true, //adds a bias parameter 
                activation: 'sigmoid', //activation function
            }));

            model.add(tf.layers.dense({
                units: 5, //number of nodes in layer
                useBias: true, //adds a bias parameter 
                activation: 'tanh', //activation function
            }));
            
            //output layer with one node
            model.add(tf.layers.dense({
                units: 1, //number of nodes in layer
                useBias: true, //adds a bias parameter 
                activation: 'tanh', //activation function
            }));

            const optimizer = tf.train.adam(0.1); //parameter = learning rate
            
            model.compile({
                loss: 'meanSquaredError',
                optimizer
            })

            return model;
        }